• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*M///////////////////////////////////////////////////////////////////////////////////////
2 //
3 //  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
4 //
5 //  By downloading, copying, installing or using the software you agree to this license.
6 //  If you do not agree to this license, do not download, install,
7 //  copy or use the software.
8 //
9 //
10 //                        Intel License Agreement
11 //                For Open Source Computer Vision Library
12 //
13 // Copyright (C) 2000, Intel Corporation, all rights reserved.
14 // Third party copyrights are property of their respective owners.
15 //
16 // Redistribution and use in source and binary forms, with or without modification,
17 // are permitted provided that the following conditions are met:
18 //
19 //   * Redistribution's of source code must retain the above copyright notice,
20 //     this list of conditions and the following disclaimer.
21 //
22 //   * Redistribution's in binary form must reproduce the above copyright notice,
23 //     this list of conditions and the following disclaimer in the documentation
24 //     and/or other materials provided with the distribution.
25 //
26 //   * The name of Intel Corporation may not be used to endorse or promote products
27 //     derived from this software without specific prior written permission.
28 //
29 // This software is provided by the copyright holders and contributors "as is" and
30 // any express or implied warranties, including, but not limited to, the implied
31 // warranties of merchantability and fitness for a particular purpose are disclaimed.
32 // In no event shall the Intel Corporation or contributors be liable for any direct,
33 // indirect, incidental, special, exemplary, or consequential damages
34 // (including, but not limited to, procurement of substitute goods or services;
35 // loss of use, data, or profits; or business interruption) however caused
36 // and on any theory of liability, whether in contract, strict liability,
37 // or tort (including negligence or otherwise) arising in any way out of
38 // the use of this software, even if advised of the possibility of such damage.
39 //
40 //M*/
41 
42 /* ////////////////////////////////////////////////////////////////////
43 //
44 //  Geometrical transforms on images and matrices: rotation, zoom etc.
45 //
46 // */
47 
48 #include "_cv.h"
49 
50 
51 /************** interpolation constants and tables ***************/
52 
53 #define ICV_WARP_MUL_ONE_8U(x)  ((x) << ICV_WARP_SHIFT)
54 #define ICV_WARP_DESCALE_8U(x)  CV_DESCALE((x), ICV_WARP_SHIFT*2)
55 #define ICV_WARP_CLIP_X(x)      ((unsigned)(x) < (unsigned)ssize.width ? \
56                                 (x) : (x) < 0 ? 0 : ssize.width - 1)
57 #define ICV_WARP_CLIP_Y(y)      ((unsigned)(y) < (unsigned)ssize.height ? \
58                                 (y) : (y) < 0 ? 0 : ssize.height - 1)
59 
60 float icvLinearCoeffs[(ICV_LINEAR_TAB_SIZE+1)*2];
61 
icvInitLinearCoeffTab()62 void icvInitLinearCoeffTab()
63 {
64     static int inittab = 0;
65     if( !inittab )
66     {
67         for( int i = 0; i <= ICV_LINEAR_TAB_SIZE; i++ )
68         {
69             float x = (float)i/ICV_LINEAR_TAB_SIZE;
70             icvLinearCoeffs[i*2] = x;
71             icvLinearCoeffs[i*2+1] = 1.f - x;
72         }
73 
74         inittab = 1;
75     }
76 }
77 
78 
79 float icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE+1)*2];
80 
icvInitCubicCoeffTab()81 void icvInitCubicCoeffTab()
82 {
83     static int inittab = 0;
84     if( !inittab )
85     {
86 #if 0
87         // classical Mitchell-Netravali filter
88         const double B = 1./3;
89         const double C = 1./3;
90         const double p0 = (6 - 2*B)/6.;
91         const double p2 = (-18 + 12*B + 6*C)/6.;
92         const double p3 = (12 - 9*B - 6*C)/6.;
93         const double q0 = (8*B + 24*C)/6.;
94         const double q1 = (-12*B - 48*C)/6.;
95         const double q2 = (6*B + 30*C)/6.;
96         const double q3 = (-B - 6*C)/6.;
97 
98         #define ICV_CUBIC_1(x)  (((x)*p3 + p2)*(x)*(x) + p0)
99         #define ICV_CUBIC_2(x)  ((((x)*q3 + q2)*(x) + q1)*(x) + q0)
100 #else
101         // alternative "sharp" filter
102         const double A = -0.75;
103         #define ICV_CUBIC_1(x)  (((A + 2)*(x) - (A + 3))*(x)*(x) + 1)
104         #define ICV_CUBIC_2(x)  (((A*(x) - 5*A)*(x) + 8*A)*(x) - 4*A)
105 #endif
106         for( int i = 0; i <= ICV_CUBIC_TAB_SIZE; i++ )
107         {
108             float x = (float)i/ICV_CUBIC_TAB_SIZE;
109             icvCubicCoeffs[i*2] = (float)ICV_CUBIC_1(x);
110             x += 1.f;
111             icvCubicCoeffs[i*2+1] = (float)ICV_CUBIC_2(x);
112         }
113 
114         inittab = 1;
115     }
116 }
117 
118 
119 /****************************************************************************************\
120 *                                         Resize                                         *
121 \****************************************************************************************/
122 
123 static CvStatus CV_STDCALL
icvResize_NN_8u_C1R(const uchar * src,int srcstep,CvSize ssize,uchar * dst,int dststep,CvSize dsize,int pix_size)124 icvResize_NN_8u_C1R( const uchar* src, int srcstep, CvSize ssize,
125                      uchar* dst, int dststep, CvSize dsize, int pix_size )
126 {
127     int* x_ofs = (int*)cvStackAlloc( dsize.width * sizeof(x_ofs[0]) );
128     int pix_size4 = pix_size / sizeof(int);
129     int x, y, t;
130 
131     for( x = 0; x < dsize.width; x++ )
132     {
133         t = (ssize.width*x*2 + MIN(ssize.width, dsize.width) - 1)/(dsize.width*2);
134         t -= t >= ssize.width;
135         x_ofs[x] = t*pix_size;
136     }
137 
138     for( y = 0; y < dsize.height; y++, dst += dststep )
139     {
140         const uchar* tsrc;
141         t = (ssize.height*y*2 + MIN(ssize.height, dsize.height) - 1)/(dsize.height*2);
142         t -= t >= ssize.height;
143         tsrc = src + srcstep*t;
144 
145         switch( pix_size )
146         {
147         case 1:
148             for( x = 0; x <= dsize.width - 2; x += 2 )
149             {
150                 uchar t0 = tsrc[x_ofs[x]];
151                 uchar t1 = tsrc[x_ofs[x+1]];
152 
153                 dst[x] = t0;
154                 dst[x+1] = t1;
155             }
156 
157             for( ; x < dsize.width; x++ )
158                 dst[x] = tsrc[x_ofs[x]];
159             break;
160         case 2:
161             for( x = 0; x < dsize.width; x++ )
162                 *(ushort*)(dst + x*2) = *(ushort*)(tsrc + x_ofs[x]);
163             break;
164         case 3:
165             for( x = 0; x < dsize.width; x++ )
166             {
167                 const uchar* _tsrc = tsrc + x_ofs[x];
168                 dst[x*3] = _tsrc[0]; dst[x*3+1] = _tsrc[1]; dst[x*3+2] = _tsrc[2];
169             }
170             break;
171         case 4:
172             for( x = 0; x < dsize.width; x++ )
173                 *(int*)(dst + x*4) = *(int*)(tsrc + x_ofs[x]);
174             break;
175         case 6:
176             for( x = 0; x < dsize.width; x++ )
177             {
178                 const ushort* _tsrc = (const ushort*)(tsrc + x_ofs[x]);
179                 ushort* _tdst = (ushort*)(dst + x*6);
180                 _tdst[0] = _tsrc[0]; _tdst[1] = _tsrc[1]; _tdst[2] = _tsrc[2];
181             }
182             break;
183         default:
184             for( x = 0; x < dsize.width; x++ )
185                 CV_MEMCPY_INT( dst + x*pix_size, tsrc + x_ofs[x], pix_size4 );
186         }
187     }
188 
189     return CV_OK;
190 }
191 
192 
193 typedef struct CvResizeAlpha
194 {
195     int idx;
196     union
197     {
198         float alpha;
199         int ialpha;
200     };
201 }
202 CvResizeAlpha;
203 
204 
205 #define  ICV_DEF_RESIZE_BILINEAR_FUNC( flavor, arrtype, worktype, alpha_field,  \
206                                        mul_one_macro, descale_macro )           \
207 static CvStatus CV_STDCALL                                                      \
208 icvResize_Bilinear_##flavor##_CnR( const arrtype* src, int srcstep, CvSize ssize,\
209                                    arrtype* dst, int dststep, CvSize dsize,     \
210                                    int cn, int xmax,                            \
211                                    const CvResizeAlpha* xofs,                   \
212                                    const CvResizeAlpha* yofs,                   \
213                                    worktype* buf0, worktype* buf1 )             \
214 {                                                                               \
215     int prev_sy0 = -1, prev_sy1 = -1;                                           \
216     int k, dx, dy;                                                              \
217                                                                                 \
218     srcstep /= sizeof(src[0]);                                                  \
219     dststep /= sizeof(dst[0]);                                                  \
220     dsize.width *= cn;                                                          \
221     xmax *= cn;                                                                 \
222                                                                                 \
223     for( dy = 0; dy < dsize.height; dy++, dst += dststep )                      \
224     {                                                                           \
225         worktype fy = yofs[dy].alpha_field, *swap_t;                            \
226         int sy0 = yofs[dy].idx, sy1 = sy0 + (fy > 0 && sy0 < ssize.height-1);   \
227                                                                                 \
228         if( sy0 == prev_sy0 && sy1 == prev_sy1 )                                \
229             k = 2;                                                              \
230         else if( sy0 == prev_sy1 )                                              \
231         {                                                                       \
232             CV_SWAP( buf0, buf1, swap_t );                                      \
233             k = 1;                                                              \
234         }                                                                       \
235         else                                                                    \
236             k = 0;                                                              \
237                                                                                 \
238         for( ; k < 2; k++ )                                                     \
239         {                                                                       \
240             worktype* _buf = k == 0 ? buf0 : buf1;                              \
241             const arrtype* _src;                                                \
242             int sy = k == 0 ? sy0 : sy1;                                        \
243             if( k == 1 && sy1 == sy0 )                                          \
244             {                                                                   \
245                 memcpy( buf1, buf0, dsize.width*sizeof(buf0[0]) );              \
246                 continue;                                                       \
247             }                                                                   \
248                                                                                 \
249             _src = src + sy*srcstep;                                            \
250             for( dx = 0; dx < xmax; dx++ )                                      \
251             {                                                                   \
252                 int sx = xofs[dx].idx;                                          \
253                 worktype fx = xofs[dx].alpha_field;                             \
254                 worktype t = _src[sx];                                          \
255                 _buf[dx] = mul_one_macro(t) + fx*(_src[sx+cn] - t);             \
256             }                                                                   \
257                                                                                 \
258             for( ; dx < dsize.width; dx++ )                                     \
259                 _buf[dx] = mul_one_macro(_src[xofs[dx].idx]);                   \
260         }                                                                       \
261                                                                                 \
262         prev_sy0 = sy0;                                                         \
263         prev_sy1 = sy1;                                                         \
264                                                                                 \
265         if( sy0 == sy1 )                                                        \
266             for( dx = 0; dx < dsize.width; dx++ )                               \
267                 dst[dx] = (arrtype)descale_macro( mul_one_macro(buf0[dx]));     \
268         else                                                                    \
269             for( dx = 0; dx < dsize.width; dx++ )                               \
270                 dst[dx] = (arrtype)descale_macro( mul_one_macro(buf0[dx]) +     \
271                                                   fy*(buf1[dx] - buf0[dx]));    \
272     }                                                                           \
273                                                                                 \
274     return CV_OK;                                                               \
275 }
276 
277 
278 typedef struct CvDecimateAlpha
279 {
280     int si, di;
281     float alpha;
282 }
283 CvDecimateAlpha;
284 
285 
286 #define  ICV_DEF_RESIZE_AREA_FAST_FUNC( flavor, arrtype, worktype, cast_macro ) \
287 static CvStatus CV_STDCALL                                                      \
288 icvResize_AreaFast_##flavor##_CnR( const arrtype* src, int srcstep, CvSize ssize,\
289                               arrtype* dst, int dststep, CvSize dsize, int cn,  \
290                               const int* ofs, const int* xofs )                 \
291 {                                                                               \
292     int dy, dx, k = 0;                                                          \
293     int scale_x = ssize.width/dsize.width;                                      \
294     int scale_y = ssize.height/dsize.height;                                    \
295     int area = scale_x*scale_y;                                                 \
296     float scale = 1.f/(scale_x*scale_y);                                        \
297                                                                                 \
298     srcstep /= sizeof(src[0]);                                                  \
299     dststep /= sizeof(dst[0]);                                                  \
300     dsize.width *= cn;                                                          \
301                                                                                 \
302     for( dy = 0; dy < dsize.height; dy++, dst += dststep )                      \
303         for( dx = 0; dx < dsize.width; dx++ )                                   \
304         {                                                                       \
305             const arrtype* _src = src + dy*scale_y*srcstep + xofs[dx];          \
306             worktype sum = 0;                                                   \
307                                                                                 \
308             for( k = 0; k <= area - 4; k += 4 )                                 \
309                 sum += _src[ofs[k]] + _src[ofs[k+1]] +                          \
310                        _src[ofs[k+2]] + _src[ofs[k+3]];                         \
311                                                                                 \
312             for( ; k < area; k++ )                                              \
313                 sum += _src[ofs[k]];                                            \
314                                                                                 \
315             dst[dx] = (arrtype)cast_macro( sum*scale );                         \
316         }                                                                       \
317                                                                                 \
318     return CV_OK;                                                               \
319 }
320 
321 
322 #define  ICV_DEF_RESIZE_AREA_FUNC( flavor, arrtype, load_macro, cast_macro )    \
323 static CvStatus CV_STDCALL                                                      \
324 icvResize_Area_##flavor##_CnR( const arrtype* src, int srcstep, CvSize ssize,   \
325                                arrtype* dst, int dststep, CvSize dsize,         \
326                                int cn, const CvDecimateAlpha* xofs,             \
327                                int xofs_count, float* buf, float* sum )         \
328 {                                                                               \
329     int k, sy, dx, cur_dy = 0;                                                  \
330     float scale_y = (float)ssize.height/dsize.height;                           \
331                                                                                 \
332     srcstep /= sizeof(src[0]);                                                  \
333     dststep /= sizeof(dst[0]);                                                  \
334     dsize.width *= cn;                                                          \
335                                                                                 \
336     for( sy = 0; sy < ssize.height; sy++, src += srcstep )                      \
337     {                                                                           \
338         if( cn == 1 )                                                           \
339             for( k = 0; k < xofs_count; k++ )                                   \
340             {                                                                   \
341                 int dxn = xofs[k].di;                                           \
342                 float alpha = xofs[k].alpha;                                    \
343                 buf[dxn] = buf[dxn] + load_macro(src[xofs[k].si])*alpha;        \
344             }                                                                   \
345         else if( cn == 2 )                                                      \
346             for( k = 0; k < xofs_count; k++ )                                   \
347             {                                                                   \
348                 int sxn = xofs[k].si;                                           \
349                 int dxn = xofs[k].di;                                           \
350                 float alpha = xofs[k].alpha;                                    \
351                 float t0 = buf[dxn] + load_macro(src[sxn])*alpha;               \
352                 float t1 = buf[dxn+1] + load_macro(src[sxn+1])*alpha;           \
353                 buf[dxn] = t0; buf[dxn+1] = t1;                                 \
354             }                                                                   \
355         else if( cn == 3 )                                                      \
356             for( k = 0; k < xofs_count; k++ )                                   \
357             {                                                                   \
358                 int sxn = xofs[k].si;                                           \
359                 int dxn = xofs[k].di;                                           \
360                 float alpha = xofs[k].alpha;                                    \
361                 float t0 = buf[dxn] + load_macro(src[sxn])*alpha;               \
362                 float t1 = buf[dxn+1] + load_macro(src[sxn+1])*alpha;           \
363                 float t2 = buf[dxn+2] + load_macro(src[sxn+2])*alpha;           \
364                 buf[dxn] = t0; buf[dxn+1] = t1; buf[dxn+2] = t2;                \
365             }                                                                   \
366         else                                                                    \
367             for( k = 0; k < xofs_count; k++ )                                   \
368             {                                                                   \
369                 int sxn = xofs[k].si;                                           \
370                 int dxn = xofs[k].di;                                           \
371                 float alpha = xofs[k].alpha;                                    \
372                 float t0 = buf[dxn] + load_macro(src[sxn])*alpha;               \
373                 float t1 = buf[dxn+1] + load_macro(src[sxn+1])*alpha;           \
374                 buf[dxn] = t0; buf[dxn+1] = t1;                                 \
375                 t0 = buf[dxn+2] + load_macro(src[sxn+2])*alpha;                 \
376                 t1 = buf[dxn+3] + load_macro(src[sxn+3])*alpha;                 \
377                 buf[dxn+2] = t0; buf[dxn+3] = t1;                               \
378             }                                                                   \
379                                                                                 \
380         if( (cur_dy + 1)*scale_y <= sy + 1 || sy == ssize.height - 1 )          \
381         {                                                                       \
382             float beta = sy + 1 - (cur_dy+1)*scale_y, beta1;                    \
383             beta = MAX( beta, 0 );                                              \
384             beta1 = 1 - beta;                                                   \
385             if( fabs(beta) < 1e-3 )                                             \
386                 for( dx = 0; dx < dsize.width; dx++ )                           \
387                 {                                                               \
388                     dst[dx] = (arrtype)cast_macro(sum[dx] + buf[dx]);           \
389                     sum[dx] = buf[dx] = 0;                                      \
390                 }                                                               \
391             else                                                                \
392                 for( dx = 0; dx < dsize.width; dx++ )                           \
393                 {                                                               \
394                     dst[dx] = (arrtype)cast_macro(sum[dx] + buf[dx]*beta1);     \
395                     sum[dx] = buf[dx]*beta;                                     \
396                     buf[dx] = 0;                                                \
397                 }                                                               \
398             dst += dststep;                                                     \
399             cur_dy++;                                                           \
400         }                                                                       \
401         else                                                                    \
402             for( dx = 0; dx < dsize.width; dx += 2 )                            \
403             {                                                                   \
404                 float t0 = sum[dx] + buf[dx];                                   \
405                 float t1 = sum[dx+1] + buf[dx+1];                               \
406                 sum[dx] = t0; sum[dx+1] = t1;                                   \
407                 buf[dx] = buf[dx+1] = 0;                                        \
408             }                                                                   \
409     }                                                                           \
410                                                                                 \
411     return CV_OK;                                                               \
412 }
413 
414 
415 #define  ICV_DEF_RESIZE_BICUBIC_FUNC( flavor, arrtype, worktype, load_macro,    \
416                                       cast_macro1, cast_macro2 )                \
417 static CvStatus CV_STDCALL                                                      \
418 icvResize_Bicubic_##flavor##_CnR( const arrtype* src, int srcstep, CvSize ssize,\
419                                   arrtype* dst, int dststep, CvSize dsize,      \
420                                   int cn, int xmin, int xmax,                   \
421                                   const CvResizeAlpha* xofs, float** buf )      \
422 {                                                                               \
423     float scale_y = (float)ssize.height/dsize.height;                           \
424     int dx, dy, sx, sy, sy2, ify;                                               \
425     int prev_sy2 = -2;                                                          \
426                                                                                 \
427     xmin *= cn; xmax *= cn;                                                     \
428     dsize.width *= cn;                                                          \
429     ssize.width *= cn;                                                          \
430     srcstep /= sizeof(src[0]);                                                  \
431     dststep /= sizeof(dst[0]);                                                  \
432                                                                                 \
433     for( dy = 0; dy < dsize.height; dy++, dst += dststep )                      \
434     {                                                                           \
435         float w0, w1, w2, w3;                                                   \
436         float fy, x, sum;                                                       \
437         float *row, *row0, *row1, *row2, *row3;                                 \
438         int k1, k = 4;                                                          \
439                                                                                 \
440         fy = dy*scale_y;                                                        \
441         sy = cvFloor(fy);                                                       \
442         fy -= sy;                                                               \
443         ify = cvRound(fy*ICV_CUBIC_TAB_SIZE);                                   \
444         sy2 = sy + 2;                                                           \
445                                                                                 \
446         if( sy2 > prev_sy2 )                                                    \
447         {                                                                       \
448             int delta = prev_sy2 - sy + 2;                                      \
449             for( k = 0; k < delta; k++ )                                        \
450                 CV_SWAP( buf[k], buf[k+4-delta], row );                         \
451         }                                                                       \
452                                                                                 \
453         for( sy += k - 1; k < 4; k++, sy++ )                                    \
454         {                                                                       \
455             const arrtype* _src = src + sy*srcstep;                             \
456                                                                                 \
457             row = buf[k];                                                       \
458             if( sy < 0 )                                                        \
459                 continue;                                                       \
460             if( sy >= ssize.height )                                            \
461             {                                                                   \
462                 assert( k > 0 );                                                \
463                 memcpy( row, buf[k-1], dsize.width*sizeof(row[0]) );            \
464                 continue;                                                       \
465             }                                                                   \
466                                                                                 \
467             for( dx = 0; dx < xmin; dx++ )                                      \
468             {                                                                   \
469                 int ifx = xofs[dx].ialpha, sx0 = xofs[dx].idx;                  \
470                 sx = sx0 + cn*2;                                                \
471                 while( sx >= ssize.width )                                      \
472                     sx -= cn;                                                   \
473                 x = load_macro(_src[sx]);                                       \
474                 sum = x*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE - ifx)*2 + 1];       \
475                 if( (unsigned)(sx = sx0 + cn) < (unsigned)ssize.width )         \
476                     x = load_macro(_src[sx]);                                   \
477                 sum += x*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE - ifx)*2];          \
478                 if( (unsigned)(sx = sx0) < (unsigned)ssize.width )              \
479                     x = load_macro(_src[sx]);                                   \
480                 sum += x*icvCubicCoeffs[ifx*2];                                 \
481                 if( (unsigned)(sx = sx0 - cn) < (unsigned)ssize.width )         \
482                     x = load_macro(_src[sx]);                                   \
483                 row[dx] = sum + x*icvCubicCoeffs[ifx*2 + 1];                    \
484             }                                                                   \
485                                                                                 \
486             for( ; dx < xmax; dx++ )                                            \
487             {                                                                   \
488                 int ifx = xofs[dx].ialpha;                                      \
489                 int sx0 = xofs[dx].idx;                                         \
490                 row[dx] = _src[sx0 - cn]*icvCubicCoeffs[ifx*2 + 1] +            \
491                     _src[sx0]*icvCubicCoeffs[ifx*2] +                           \
492                     _src[sx0 + cn]*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2] + \
493                     _src[sx0 + cn*2]*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2+1];\
494             }                                                                   \
495                                                                                 \
496             for( ; dx < dsize.width; dx++ )                                     \
497             {                                                                   \
498                 int ifx = xofs[dx].ialpha, sx0 = xofs[dx].idx;                  \
499                 x = load_macro(_src[sx0 - cn]);                                 \
500                 sum = x*icvCubicCoeffs[ifx*2 + 1];                              \
501                 if( (unsigned)(sx = sx0) < (unsigned)ssize.width )              \
502                     x = load_macro(_src[sx]);                                   \
503                 sum += x*icvCubicCoeffs[ifx*2];                                 \
504                 if( (unsigned)(sx = sx0 + cn) < (unsigned)ssize.width )         \
505                     x = load_macro(_src[sx]);                                   \
506                 sum += x*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE - ifx)*2];          \
507                 if( (unsigned)(sx = sx0 + cn*2) < (unsigned)ssize.width )       \
508                     x = load_macro(_src[sx]);                                   \
509                 row[dx] = sum + x*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2+1]; \
510             }                                                                   \
511                                                                                 \
512             if( sy == 0 )                                                       \
513                 for( k1 = 0; k1 < k; k1++ )                                     \
514                     memcpy( buf[k1], row, dsize.width*sizeof(row[0]));          \
515         }                                                                       \
516                                                                                 \
517         prev_sy2 = sy2;                                                         \
518                                                                                 \
519         row0 = buf[0]; row1 = buf[1];                                           \
520         row2 = buf[2]; row3 = buf[3];                                           \
521                                                                                 \
522         w0 = icvCubicCoeffs[ify*2+1];                                           \
523         w1 = icvCubicCoeffs[ify*2];                                             \
524         w2 = icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE - ify)*2];                      \
525         w3 = icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE - ify)*2 + 1];                  \
526                                                                                 \
527         for( dx = 0; dx < dsize.width; dx++ )                                   \
528         {                                                                       \
529             worktype val = cast_macro1( row0[dx]*w0 + row1[dx]*w1 +             \
530                                         row2[dx]*w2 + row3[dx]*w3 );            \
531             dst[dx] = cast_macro2(val);                                         \
532         }                                                                       \
533     }                                                                           \
534                                                                                 \
535     return CV_OK;                                                               \
536 }
537 
538 
539 ICV_DEF_RESIZE_BILINEAR_FUNC( 8u, uchar, int, ialpha,
540                               ICV_WARP_MUL_ONE_8U, ICV_WARP_DESCALE_8U )
541 ICV_DEF_RESIZE_BILINEAR_FUNC( 16u, ushort, float, alpha, CV_NOP, cvRound )
542 ICV_DEF_RESIZE_BILINEAR_FUNC( 32f, float, float, alpha, CV_NOP, CV_NOP )
543 
544 ICV_DEF_RESIZE_BICUBIC_FUNC( 8u, uchar, int, CV_8TO32F, cvRound, CV_CAST_8U )
545 ICV_DEF_RESIZE_BICUBIC_FUNC( 16u, ushort, int, CV_NOP, cvRound, CV_CAST_16U )
546 ICV_DEF_RESIZE_BICUBIC_FUNC( 32f, float, float, CV_NOP, CV_NOP, CV_NOP )
547 
548 ICV_DEF_RESIZE_AREA_FAST_FUNC( 8u, uchar, int, cvRound )
549 ICV_DEF_RESIZE_AREA_FAST_FUNC( 16u, ushort, int, cvRound )
550 ICV_DEF_RESIZE_AREA_FAST_FUNC( 32f, float, float, CV_NOP )
551 
552 ICV_DEF_RESIZE_AREA_FUNC( 8u, uchar, CV_8TO32F, cvRound )
553 ICV_DEF_RESIZE_AREA_FUNC( 16u, ushort, CV_NOP, cvRound )
554 ICV_DEF_RESIZE_AREA_FUNC( 32f, float, CV_NOP, CV_NOP )
555 
556 
icvInitResizeTab(CvFuncTable * bilin_tab,CvFuncTable * bicube_tab,CvFuncTable * areafast_tab,CvFuncTable * area_tab)557 static void icvInitResizeTab( CvFuncTable* bilin_tab,
558                               CvFuncTable* bicube_tab,
559                               CvFuncTable* areafast_tab,
560                               CvFuncTable* area_tab )
561 {
562     bilin_tab->fn_2d[CV_8U] = (void*)icvResize_Bilinear_8u_CnR;
563     bilin_tab->fn_2d[CV_16U] = (void*)icvResize_Bilinear_16u_CnR;
564     bilin_tab->fn_2d[CV_32F] = (void*)icvResize_Bilinear_32f_CnR;
565 
566     bicube_tab->fn_2d[CV_8U] = (void*)icvResize_Bicubic_8u_CnR;
567     bicube_tab->fn_2d[CV_16U] = (void*)icvResize_Bicubic_16u_CnR;
568     bicube_tab->fn_2d[CV_32F] = (void*)icvResize_Bicubic_32f_CnR;
569 
570     areafast_tab->fn_2d[CV_8U] = (void*)icvResize_AreaFast_8u_CnR;
571     areafast_tab->fn_2d[CV_16U] = (void*)icvResize_AreaFast_16u_CnR;
572     areafast_tab->fn_2d[CV_32F] = (void*)icvResize_AreaFast_32f_CnR;
573 
574     area_tab->fn_2d[CV_8U] = (void*)icvResize_Area_8u_CnR;
575     area_tab->fn_2d[CV_16U] = (void*)icvResize_Area_16u_CnR;
576     area_tab->fn_2d[CV_32F] = (void*)icvResize_Area_32f_CnR;
577 }
578 
579 
580 typedef CvStatus (CV_STDCALL * CvResizeBilinearFunc)
581                     ( const void* src, int srcstep, CvSize ssize,
582                       void* dst, int dststep, CvSize dsize,
583                       int cn, int xmax, const CvResizeAlpha* xofs,
584                       const CvResizeAlpha* yofs, float* buf0, float* buf1 );
585 
586 typedef CvStatus (CV_STDCALL * CvResizeBicubicFunc)
587                     ( const void* src, int srcstep, CvSize ssize,
588                       void* dst, int dststep, CvSize dsize,
589                       int cn, int xmin, int xmax,
590                       const CvResizeAlpha* xofs, float** buf );
591 
592 typedef CvStatus (CV_STDCALL * CvResizeAreaFastFunc)
593                     ( const void* src, int srcstep, CvSize ssize,
594                       void* dst, int dststep, CvSize dsize,
595                       int cn, const int* ofs, const int *xofs );
596 
597 typedef CvStatus (CV_STDCALL * CvResizeAreaFunc)
598                     ( const void* src, int srcstep, CvSize ssize,
599                       void* dst, int dststep, CvSize dsize,
600                       int cn, const CvDecimateAlpha* xofs,
601                       int xofs_count, float* buf, float* sum );
602 
603 
604 ////////////////////////////////// IPP resize functions //////////////////////////////////
605 
606 icvResize_8u_C1R_t icvResize_8u_C1R_p = 0;
607 icvResize_8u_C3R_t icvResize_8u_C3R_p = 0;
608 icvResize_8u_C4R_t icvResize_8u_C4R_p = 0;
609 icvResize_16u_C1R_t icvResize_16u_C1R_p = 0;
610 icvResize_16u_C3R_t icvResize_16u_C3R_p = 0;
611 icvResize_16u_C4R_t icvResize_16u_C4R_p = 0;
612 icvResize_32f_C1R_t icvResize_32f_C1R_p = 0;
613 icvResize_32f_C3R_t icvResize_32f_C3R_p = 0;
614 icvResize_32f_C4R_t icvResize_32f_C4R_p = 0;
615 
616 typedef CvStatus (CV_STDCALL * CvResizeIPPFunc)
617 ( const void* src, CvSize srcsize, int srcstep, CvRect srcroi,
618   void* dst, int dststep, CvSize dstroi,
619   double xfactor, double yfactor, int interpolation );
620 
621 //////////////////////////////////////////////////////////////////////////////////////////
622 
623 CV_IMPL void
cvResize(const CvArr * srcarr,CvArr * dstarr,int method)624 cvResize( const CvArr* srcarr, CvArr* dstarr, int method )
625 {
626     static CvFuncTable bilin_tab, bicube_tab, areafast_tab, area_tab;
627     static int inittab = 0;
628     void* temp_buf = 0;
629 
630     CV_FUNCNAME( "cvResize" );
631 
632     __BEGIN__;
633 
634     CvMat srcstub, *src = (CvMat*)srcarr;
635     CvMat dststub, *dst = (CvMat*)dstarr;
636     CvSize ssize, dsize;
637     float scale_x, scale_y;
638     int k, sx, sy, dx, dy;
639     int type, depth, cn;
640 
641     CV_CALL( src = cvGetMat( srcarr, &srcstub ));
642     CV_CALL( dst = cvGetMat( dstarr, &dststub ));
643 
644     if( CV_ARE_SIZES_EQ( src, dst ))
645     {
646         CV_CALL( cvCopy( src, dst ));
647         EXIT;
648     }
649 
650     if( !CV_ARE_TYPES_EQ( src, dst ))
651         CV_ERROR( CV_StsUnmatchedFormats, "" );
652 
653     if( !inittab )
654     {
655         icvInitResizeTab( &bilin_tab, &bicube_tab, &areafast_tab, &area_tab );
656         inittab = 1;
657     }
658 
659     ssize = cvGetMatSize( src );
660     dsize = cvGetMatSize( dst );
661     type = CV_MAT_TYPE(src->type);
662     depth = CV_MAT_DEPTH(type);
663     cn = CV_MAT_CN(type);
664     scale_x = (float)ssize.width/dsize.width;
665     scale_y = (float)ssize.height/dsize.height;
666 
667     if( method == CV_INTER_CUBIC &&
668         (MIN(ssize.width, dsize.width) <= 4 ||
669         MIN(ssize.height, dsize.height) <= 4) )
670         method = CV_INTER_LINEAR;
671 
672     if( icvResize_8u_C1R_p &&
673         MIN(ssize.width, dsize.width) > 4 &&
674         MIN(ssize.height, dsize.height) > 4 )
675     {
676         CvResizeIPPFunc ipp_func =
677             type == CV_8UC1 ? icvResize_8u_C1R_p :
678             type == CV_8UC3 ? icvResize_8u_C3R_p :
679             type == CV_8UC4 ? icvResize_8u_C4R_p :
680             type == CV_16UC1 ? icvResize_16u_C1R_p :
681             type == CV_16UC3 ? icvResize_16u_C3R_p :
682             type == CV_16UC4 ? icvResize_16u_C4R_p :
683             type == CV_32FC1 ? icvResize_32f_C1R_p :
684             type == CV_32FC3 ? icvResize_32f_C3R_p :
685             type == CV_32FC4 ? icvResize_32f_C4R_p : 0;
686         if( ipp_func && (CV_INTER_NN < method && method < CV_INTER_AREA))
687         {
688             int srcstep = src->step ? src->step : CV_STUB_STEP;
689             int dststep = dst->step ? dst->step : CV_STUB_STEP;
690             IPPI_CALL( ipp_func( src->data.ptr, ssize, srcstep,
691                                  cvRect(0,0,ssize.width,ssize.height),
692                                  dst->data.ptr, dststep, dsize,
693                                  (double)dsize.width/ssize.width,
694                                  (double)dsize.height/ssize.height, 1 << method ));
695             EXIT;
696         }
697     }
698 
699     if( method == CV_INTER_NN )
700     {
701         IPPI_CALL( icvResize_NN_8u_C1R( src->data.ptr, src->step, ssize,
702                                         dst->data.ptr, dst->step, dsize,
703                                         CV_ELEM_SIZE(src->type)));
704     }
705     else if( method == CV_INTER_LINEAR || method == CV_INTER_AREA )
706     {
707         if( method == CV_INTER_AREA &&
708             ssize.width >= dsize.width && ssize.height >= dsize.height )
709         {
710             // "area" method for (scale_x > 1 & scale_y > 1)
711             int iscale_x = cvRound(scale_x);
712             int iscale_y = cvRound(scale_y);
713 
714             if( fabs(scale_x - iscale_x) < DBL_EPSILON &&
715                 fabs(scale_y - iscale_y) < DBL_EPSILON )
716             {
717                 int area = iscale_x*iscale_y;
718                 int srcstep = src->step / CV_ELEM_SIZE(depth);
719                 int* ofs = (int*)cvStackAlloc( (area + dsize.width*cn)*sizeof(int) );
720                 int* xofs = ofs + area;
721                 CvResizeAreaFastFunc func = (CvResizeAreaFastFunc)areafast_tab.fn_2d[depth];
722 
723                 if( !func )
724                     CV_ERROR( CV_StsUnsupportedFormat, "" );
725 
726                 for( sy = 0, k = 0; sy < iscale_y; sy++ )
727                     for( sx = 0; sx < iscale_x; sx++ )
728                         ofs[k++] = sy*srcstep + sx*cn;
729 
730                 for( dx = 0; dx < dsize.width; dx++ )
731                 {
732                     sx = dx*iscale_x*cn;
733                     for( k = 0; k < cn; k++ )
734                         xofs[dx*cn + k] = sx + k;
735                 }
736 
737                 IPPI_CALL( func( src->data.ptr, src->step, ssize, dst->data.ptr,
738                                  dst->step, dsize, cn, ofs, xofs ));
739             }
740             else
741             {
742                 int buf_len = dsize.width*cn + 4, buf_size, xofs_count = 0;
743                 float scale = 1.f/(scale_x*scale_y);
744                 float *buf, *sum;
745                 CvDecimateAlpha* xofs;
746                 CvResizeAreaFunc func = (CvResizeAreaFunc)area_tab.fn_2d[depth];
747 
748                 if( !func || cn > 4 )
749                     CV_ERROR( CV_StsUnsupportedFormat, "" );
750 
751                 buf_size = buf_len*2*sizeof(float) + ssize.width*2*sizeof(CvDecimateAlpha);
752                 if( buf_size < CV_MAX_LOCAL_SIZE )
753                     buf = (float*)cvStackAlloc(buf_size);
754                 else
755                     CV_CALL( temp_buf = buf = (float*)cvAlloc(buf_size));
756                 sum = buf + buf_len;
757                 xofs = (CvDecimateAlpha*)(sum + buf_len);
758 
759                 for( dx = 0, k = 0; dx < dsize.width; dx++ )
760                 {
761                     float fsx1 = dx*scale_x, fsx2 = fsx1 + scale_x;
762                     int sx1 = cvCeil(fsx1), sx2 = cvFloor(fsx2);
763 
764                     assert( (unsigned)sx1 < (unsigned)ssize.width );
765 
766                     if( sx1 > fsx1 )
767                     {
768                         assert( k < ssize.width*2 );
769                         xofs[k].di = dx*cn;
770                         xofs[k].si = (sx1-1)*cn;
771                         xofs[k++].alpha = (sx1 - fsx1)*scale;
772                     }
773 
774                     for( sx = sx1; sx < sx2; sx++ )
775                     {
776                         assert( k < ssize.width*2 );
777                         xofs[k].di = dx*cn;
778                         xofs[k].si = sx*cn;
779                         xofs[k++].alpha = scale;
780                     }
781 
782                     if( fsx2 - sx2 > 1e-3 )
783                     {
784                         assert( k < ssize.width*2 );
785                         assert((unsigned)sx2 < (unsigned)ssize.width );
786                         xofs[k].di = dx*cn;
787                         xofs[k].si = sx2*cn;
788                         xofs[k++].alpha = (fsx2 - sx2)*scale;
789                     }
790                 }
791 
792                 xofs_count = k;
793                 memset( sum, 0, buf_len*sizeof(float) );
794                 memset( buf, 0, buf_len*sizeof(float) );
795 
796                 IPPI_CALL( func( src->data.ptr, src->step, ssize, dst->data.ptr,
797                                  dst->step, dsize, cn, xofs, xofs_count, buf, sum ));
798             }
799         }
800         else // true "area" method for the cases (scale_x > 1 & scale_y < 1) and
801              // (scale_x < 1 & scale_y > 1) is not implemented.
802              // instead, it is emulated via some variant of bilinear interpolation.
803         {
804             float inv_scale_x = (float)dsize.width/ssize.width;
805             float inv_scale_y = (float)dsize.height/ssize.height;
806             int xmax = dsize.width, width = dsize.width*cn, buf_size;
807             float *buf0, *buf1;
808             CvResizeAlpha *xofs, *yofs;
809             int area_mode = method == CV_INTER_AREA;
810             float fx, fy;
811             CvResizeBilinearFunc func = (CvResizeBilinearFunc)bilin_tab.fn_2d[depth];
812 
813             if( !func )
814                 CV_ERROR( CV_StsUnsupportedFormat, "" );
815 
816             buf_size = width*2*sizeof(float) + (width + dsize.height)*sizeof(CvResizeAlpha);
817             if( buf_size < CV_MAX_LOCAL_SIZE )
818                 buf0 = (float*)cvStackAlloc(buf_size);
819             else
820                 CV_CALL( temp_buf = buf0 = (float*)cvAlloc(buf_size));
821             buf1 = buf0 + width;
822             xofs = (CvResizeAlpha*)(buf1 + width);
823             yofs = xofs + width;
824 
825             for( dx = 0; dx < dsize.width; dx++ )
826             {
827                 if( !area_mode )
828                 {
829                     fx = (float)((dx+0.5)*scale_x - 0.5);
830                     sx = cvFloor(fx);
831                     fx -= sx;
832                 }
833                 else
834                 {
835                     sx = cvFloor(dx*scale_x);
836                     fx = (dx+1) - (sx+1)*inv_scale_x;
837                     fx = fx <= 0 ? 0.f : fx - cvFloor(fx);
838                 }
839 
840                 if( sx < 0 )
841                     fx = 0, sx = 0;
842 
843                 if( sx >= ssize.width-1 )
844                 {
845                     fx = 0, sx = ssize.width-1;
846                     if( xmax >= dsize.width )
847                         xmax = dx;
848                 }
849 
850                 if( depth != CV_8U )
851                     for( k = 0, sx *= cn; k < cn; k++ )
852                         xofs[dx*cn + k].idx = sx + k, xofs[dx*cn + k].alpha = fx;
853                 else
854                     for( k = 0, sx *= cn; k < cn; k++ )
855                         xofs[dx*cn + k].idx = sx + k,
856                         xofs[dx*cn + k].ialpha = CV_FLT_TO_FIX(fx, ICV_WARP_SHIFT);
857             }
858 
859             for( dy = 0; dy < dsize.height; dy++ )
860             {
861                 if( !area_mode )
862                 {
863                     fy = (float)((dy+0.5)*scale_y - 0.5);
864                     sy = cvFloor(fy);
865                     fy -= sy;
866                     if( sy < 0 )
867                         sy = 0, fy = 0;
868                 }
869                 else
870                 {
871                     sy = cvFloor(dy*scale_y);
872                     fy = (dy+1) - (sy+1)*inv_scale_y;
873                     fy = fy <= 0 ? 0.f : fy - cvFloor(fy);
874                 }
875 
876                 yofs[dy].idx = sy;
877                 if( depth != CV_8U )
878                     yofs[dy].alpha = fy;
879                 else
880                     yofs[dy].ialpha = CV_FLT_TO_FIX(fy, ICV_WARP_SHIFT);
881             }
882 
883             IPPI_CALL( func( src->data.ptr, src->step, ssize, dst->data.ptr,
884                              dst->step, dsize, cn, xmax, xofs, yofs, buf0, buf1 ));
885         }
886     }
887     else if( method == CV_INTER_CUBIC )
888     {
889         int width = dsize.width*cn, buf_size;
890         int xmin = dsize.width, xmax = -1;
891         CvResizeAlpha* xofs;
892         float* buf[4];
893         CvResizeBicubicFunc func = (CvResizeBicubicFunc)bicube_tab.fn_2d[depth];
894 
895         if( !func )
896             CV_ERROR( CV_StsUnsupportedFormat, "" );
897 
898         buf_size = width*(4*sizeof(float) + sizeof(xofs[0]));
899         if( buf_size < CV_MAX_LOCAL_SIZE )
900             buf[0] = (float*)cvStackAlloc(buf_size);
901         else
902             CV_CALL( temp_buf = buf[0] = (float*)cvAlloc(buf_size));
903 
904         for( k = 1; k < 4; k++ )
905             buf[k] = buf[k-1] + width;
906         xofs = (CvResizeAlpha*)(buf[3] + width);
907 
908         icvInitCubicCoeffTab();
909 
910         for( dx = 0; dx < dsize.width; dx++ )
911         {
912             float fx = dx*scale_x;
913             sx = cvFloor(fx);
914             fx -= sx;
915             int ifx = cvRound(fx*ICV_CUBIC_TAB_SIZE);
916             if( sx-1 >= 0 && xmin > dx )
917                 xmin = dx;
918             if( sx+2 < ssize.width )
919                 xmax = dx + 1;
920 
921             // at least one of 4 points should be within the image - to
922             // be able to set other points to the same value. see the loops
923             // for( dx = 0; dx < xmin; dx++ ) ... and for( ; dx < width; dx++ ) ...
924             if( sx < -2 )
925                 sx = -2;
926             else if( sx > ssize.width )
927                 sx = ssize.width;
928 
929             for( k = 0; k < cn; k++ )
930             {
931                 xofs[dx*cn + k].idx = sx*cn + k;
932                 xofs[dx*cn + k].ialpha = ifx;
933             }
934         }
935 
936         IPPI_CALL( func( src->data.ptr, src->step, ssize, dst->data.ptr,
937                          dst->step, dsize, cn, xmin, xmax, xofs, buf ));
938     }
939     else
940         CV_ERROR( CV_StsBadFlag, "Unknown/unsupported interpolation method" );
941 
942     __END__;
943 
944     cvFree( &temp_buf );
945 }
946 
947 
948 /****************************************************************************************\
949 *                                     WarpAffine                                         *
950 \****************************************************************************************/
951 
952 #define ICV_DEF_WARP_AFFINE_BILINEAR_FUNC( flavor, arrtype, worktype,       \
953             scale_alpha_macro, mul_one_macro, descale_macro, cast_macro )   \
954 static CvStatus CV_STDCALL                                                  \
955 icvWarpAffine_Bilinear_##flavor##_CnR(                                      \
956     const arrtype* src, int step, CvSize ssize,                             \
957     arrtype* dst, int dststep, CvSize dsize,                                \
958     const double* matrix, int cn,                                           \
959     const arrtype* fillval, const int* ofs )                                \
960 {                                                                           \
961     int x, y, k;                                                            \
962     double  A12 = matrix[1], b1 = matrix[2];                                \
963     double  A22 = matrix[4], b2 = matrix[5];                                \
964                                                                             \
965     step /= sizeof(src[0]);                                                 \
966     dststep /= sizeof(dst[0]);                                              \
967                                                                             \
968     for( y = 0; y < dsize.height; y++, dst += dststep )                     \
969     {                                                                       \
970         int xs = CV_FLT_TO_FIX( A12*y + b1, ICV_WARP_SHIFT );               \
971         int ys = CV_FLT_TO_FIX( A22*y + b2, ICV_WARP_SHIFT );               \
972                                                                             \
973         for( x = 0; x < dsize.width; x++ )                                  \
974         {                                                                   \
975             int ixs = xs + ofs[x*2];                                        \
976             int iys = ys + ofs[x*2+1];                                      \
977             worktype a = scale_alpha_macro( ixs & ICV_WARP_MASK );          \
978             worktype b = scale_alpha_macro( iys & ICV_WARP_MASK );          \
979             worktype p0, p1;                                                \
980             ixs >>= ICV_WARP_SHIFT;                                         \
981             iys >>= ICV_WARP_SHIFT;                                         \
982                                                                             \
983             if( (unsigned)ixs < (unsigned)(ssize.width - 1) &&              \
984                 (unsigned)iys < (unsigned)(ssize.height - 1) )              \
985             {                                                               \
986                 const arrtype* ptr = src + step*iys + ixs*cn;               \
987                                                                             \
988                 for( k = 0; k < cn; k++ )                                   \
989                 {                                                           \
990                     p0 = mul_one_macro(ptr[k]) +                            \
991                         a * (ptr[k+cn] - ptr[k]);                           \
992                     p1 = mul_one_macro(ptr[k+step]) +                       \
993                         a * (ptr[k+cn+step] - ptr[k+step]);                 \
994                     p0 = descale_macro(mul_one_macro(p0) + b*(p1 - p0));    \
995                     dst[x*cn+k] = (arrtype)cast_macro(p0);                  \
996                 }                                                           \
997             }                                                               \
998             else if( (unsigned)(ixs+1) < (unsigned)(ssize.width+1) &&       \
999                      (unsigned)(iys+1) < (unsigned)(ssize.height+1))        \
1000             {                                                               \
1001                 int x0 = ICV_WARP_CLIP_X( ixs );                            \
1002                 int y0 = ICV_WARP_CLIP_Y( iys );                            \
1003                 int x1 = ICV_WARP_CLIP_X( ixs + 1 );                        \
1004                 int y1 = ICV_WARP_CLIP_Y( iys + 1 );                        \
1005                 const arrtype* ptr0, *ptr1, *ptr2, *ptr3;                   \
1006                                                                             \
1007                 ptr0 = src + y0*step + x0*cn;                               \
1008                 ptr1 = src + y0*step + x1*cn;                               \
1009                 ptr2 = src + y1*step + x0*cn;                               \
1010                 ptr3 = src + y1*step + x1*cn;                               \
1011                                                                             \
1012                 for( k = 0; k < cn; k++ )                                   \
1013                 {                                                           \
1014                     p0 = mul_one_macro(ptr0[k]) + a * (ptr1[k] - ptr0[k]);  \
1015                     p1 = mul_one_macro(ptr2[k]) + a * (ptr3[k] - ptr2[k]);  \
1016                     p0 = descale_macro( mul_one_macro(p0) + b*(p1 - p0) );  \
1017                     dst[x*cn+k] = (arrtype)cast_macro(p0);                  \
1018                 }                                                           \
1019             }                                                               \
1020             else if( fillval )                                              \
1021                 for( k = 0; k < cn; k++ )                                   \
1022                     dst[x*cn+k] = fillval[k];                               \
1023         }                                                                   \
1024     }                                                                       \
1025                                                                             \
1026     return CV_OK;                                                           \
1027 }
1028 
1029 
1030 #define ICV_WARP_SCALE_ALPHA(x) ((x)*(1./(ICV_WARP_MASK+1)))
1031 
1032 ICV_DEF_WARP_AFFINE_BILINEAR_FUNC( 8u, uchar, int, CV_NOP, ICV_WARP_MUL_ONE_8U,
1033                                    ICV_WARP_DESCALE_8U, CV_NOP )
1034 //ICV_DEF_WARP_AFFINE_BILINEAR_FUNC( 8u, uchar, double, ICV_WARP_SCALE_ALPHA, CV_NOP,
1035 //                                   CV_NOP, ICV_WARP_CAST_8U )
1036 ICV_DEF_WARP_AFFINE_BILINEAR_FUNC( 16u, ushort, double, ICV_WARP_SCALE_ALPHA, CV_NOP,
1037                                    CV_NOP, cvRound )
1038 ICV_DEF_WARP_AFFINE_BILINEAR_FUNC( 32f, float, double, ICV_WARP_SCALE_ALPHA, CV_NOP,
1039                                    CV_NOP, CV_NOP )
1040 
1041 
1042 typedef CvStatus (CV_STDCALL * CvWarpAffineFunc)(
1043     const void* src, int srcstep, CvSize ssize,
1044     void* dst, int dststep, CvSize dsize,
1045     const double* matrix, int cn,
1046     const void* fillval, const int* ofs );
1047 
icvInitWarpAffineTab(CvFuncTable * bilin_tab)1048 static void icvInitWarpAffineTab( CvFuncTable* bilin_tab )
1049 {
1050     bilin_tab->fn_2d[CV_8U] = (void*)icvWarpAffine_Bilinear_8u_CnR;
1051     bilin_tab->fn_2d[CV_16U] = (void*)icvWarpAffine_Bilinear_16u_CnR;
1052     bilin_tab->fn_2d[CV_32F] = (void*)icvWarpAffine_Bilinear_32f_CnR;
1053 }
1054 
1055 
1056 /////////////////////////////// IPP warpaffine functions /////////////////////////////////
1057 
1058 icvWarpAffineBack_8u_C1R_t icvWarpAffineBack_8u_C1R_p = 0;
1059 icvWarpAffineBack_8u_C3R_t icvWarpAffineBack_8u_C3R_p = 0;
1060 icvWarpAffineBack_8u_C4R_t icvWarpAffineBack_8u_C4R_p = 0;
1061 icvWarpAffineBack_32f_C1R_t icvWarpAffineBack_32f_C1R_p = 0;
1062 icvWarpAffineBack_32f_C3R_t icvWarpAffineBack_32f_C3R_p = 0;
1063 icvWarpAffineBack_32f_C4R_t icvWarpAffineBack_32f_C4R_p = 0;
1064 
1065 typedef CvStatus (CV_STDCALL * CvWarpAffineBackIPPFunc)
1066 ( const void* src, CvSize srcsize, int srcstep, CvRect srcroi,
1067   void* dst, int dststep, CvRect dstroi,
1068   const double* coeffs, int interpolation );
1069 
1070 //////////////////////////////////////////////////////////////////////////////////////////
1071 
1072 CV_IMPL void
cvWarpAffine(const CvArr * srcarr,CvArr * dstarr,const CvMat * matrix,int flags,CvScalar fillval)1073 cvWarpAffine( const CvArr* srcarr, CvArr* dstarr, const CvMat* matrix,
1074               int flags, CvScalar fillval )
1075 {
1076     static CvFuncTable bilin_tab;
1077     static int inittab = 0;
1078 
1079     CV_FUNCNAME( "cvWarpAffine" );
1080 
1081     __BEGIN__;
1082 
1083     CvMat srcstub, *src = (CvMat*)srcarr;
1084     CvMat dststub, *dst = (CvMat*)dstarr;
1085     int k, type, depth, cn, *ofs = 0;
1086     double src_matrix[6], dst_matrix[6];
1087     double fillbuf[4];
1088     int method = flags & 3;
1089     CvMat srcAb = cvMat( 2, 3, CV_64F, src_matrix ),
1090           dstAb = cvMat( 2, 3, CV_64F, dst_matrix ),
1091           A, b, invA, invAb;
1092     CvWarpAffineFunc func;
1093     CvSize ssize, dsize;
1094 
1095     if( !inittab )
1096     {
1097         icvInitWarpAffineTab( &bilin_tab );
1098         inittab = 1;
1099     }
1100 
1101     CV_CALL( src = cvGetMat( srcarr, &srcstub ));
1102     CV_CALL( dst = cvGetMat( dstarr, &dststub ));
1103 
1104     if( !CV_ARE_TYPES_EQ( src, dst ))
1105         CV_ERROR( CV_StsUnmatchedFormats, "" );
1106 
1107     if( !CV_IS_MAT(matrix) || CV_MAT_CN(matrix->type) != 1 ||
1108         CV_MAT_DEPTH(matrix->type) < CV_32F || matrix->rows != 2 || matrix->cols != 3 )
1109         CV_ERROR( CV_StsBadArg,
1110         "Transformation matrix should be 2x3 floating-point single-channel matrix" );
1111 
1112     if( flags & CV_WARP_INVERSE_MAP )
1113         cvConvertScale( matrix, &dstAb );
1114     else
1115     {
1116         // [R|t] -> [R^-1 | -(R^-1)*t]
1117         cvConvertScale( matrix, &srcAb );
1118         cvGetCols( &srcAb, &A, 0, 2 );
1119         cvGetCol( &srcAb, &b, 2 );
1120         cvGetCols( &dstAb, &invA, 0, 2 );
1121         cvGetCol( &dstAb, &invAb, 2 );
1122         cvInvert( &A, &invA, CV_SVD );
1123         cvGEMM( &invA, &b, -1, 0, 0, &invAb );
1124     }
1125 
1126     type = CV_MAT_TYPE(src->type);
1127     depth = CV_MAT_DEPTH(type);
1128     cn = CV_MAT_CN(type);
1129     if( cn > 4 )
1130         CV_ERROR( CV_BadNumChannels, "" );
1131 
1132     ssize = cvGetMatSize(src);
1133     dsize = cvGetMatSize(dst);
1134 
1135     if( icvWarpAffineBack_8u_C1R_p && MIN( ssize.width, dsize.width ) >= 4 &&
1136         MIN( ssize.height, dsize.height ) >= 4 )
1137     {
1138         CvWarpAffineBackIPPFunc ipp_func =
1139             type == CV_8UC1 ? icvWarpAffineBack_8u_C1R_p :
1140             type == CV_8UC3 ? icvWarpAffineBack_8u_C3R_p :
1141             type == CV_8UC4 ? icvWarpAffineBack_8u_C4R_p :
1142             type == CV_32FC1 ? icvWarpAffineBack_32f_C1R_p :
1143             type == CV_32FC3 ? icvWarpAffineBack_32f_C3R_p :
1144             type == CV_32FC4 ? icvWarpAffineBack_32f_C4R_p : 0;
1145 
1146         if( ipp_func && CV_INTER_NN <= method && method <= CV_INTER_AREA )
1147         {
1148             int srcstep = src->step ? src->step : CV_STUB_STEP;
1149             int dststep = dst->step ? dst->step : CV_STUB_STEP;
1150             CvRect srcroi = {0, 0, ssize.width, ssize.height};
1151             CvRect dstroi = {0, 0, dsize.width, dsize.height};
1152 
1153             // this is not the most efficient way to fill outliers
1154             if( flags & CV_WARP_FILL_OUTLIERS )
1155                 cvSet( dst, fillval );
1156 
1157             if( ipp_func( src->data.ptr, ssize, srcstep, srcroi,
1158                           dst->data.ptr, dststep, dstroi,
1159                           dstAb.data.db, 1 << method ) >= 0 )
1160                 EXIT;
1161         }
1162     }
1163 
1164     cvScalarToRawData( &fillval, fillbuf, CV_MAT_TYPE(src->type), 0 );
1165     ofs = (int*)cvStackAlloc( dst->cols*2*sizeof(ofs[0]) );
1166     for( k = 0; k < dst->cols; k++ )
1167     {
1168         ofs[2*k] = CV_FLT_TO_FIX( dst_matrix[0]*k, ICV_WARP_SHIFT );
1169         ofs[2*k+1] = CV_FLT_TO_FIX( dst_matrix[3]*k, ICV_WARP_SHIFT );
1170     }
1171 
1172     /*if( method == CV_INTER_LINEAR )*/
1173     {
1174         func = (CvWarpAffineFunc)bilin_tab.fn_2d[depth];
1175         if( !func )
1176             CV_ERROR( CV_StsUnsupportedFormat, "" );
1177 
1178         IPPI_CALL( func( src->data.ptr, src->step, ssize, dst->data.ptr,
1179                          dst->step, dsize, dst_matrix, cn,
1180                          flags & CV_WARP_FILL_OUTLIERS ? fillbuf : 0, ofs ));
1181     }
1182 
1183     __END__;
1184 }
1185 
1186 
1187 CV_IMPL CvMat*
cv2DRotationMatrix(CvPoint2D32f center,double angle,double scale,CvMat * matrix)1188 cv2DRotationMatrix( CvPoint2D32f center, double angle,
1189                     double scale, CvMat* matrix )
1190 {
1191     CV_FUNCNAME( "cvGetRotationMatrix" );
1192 
1193     __BEGIN__;
1194 
1195     double m[2][3];
1196     CvMat M = cvMat( 2, 3, CV_64FC1, m );
1197     double alpha, beta;
1198 
1199     if( !matrix )
1200         CV_ERROR( CV_StsNullPtr, "" );
1201 
1202     angle *= CV_PI/180;
1203     alpha = cos(angle)*scale;
1204     beta = sin(angle)*scale;
1205 
1206     m[0][0] = alpha;
1207     m[0][1] = beta;
1208     m[0][2] = (1-alpha)*center.x - beta*center.y;
1209     m[1][0] = -beta;
1210     m[1][1] = alpha;
1211     m[1][2] = beta*center.x + (1-alpha)*center.y;
1212 
1213     cvConvert( &M, matrix );
1214 
1215     __END__;
1216 
1217     return matrix;
1218 }
1219 
1220 
1221 /****************************************************************************************\
1222 *                                    WarpPerspective                                     *
1223 \****************************************************************************************/
1224 
1225 #define ICV_DEF_WARP_PERSPECTIVE_BILINEAR_FUNC( flavor, arrtype, load_macro, cast_macro )\
1226 static CvStatus CV_STDCALL                                                  \
1227 icvWarpPerspective_Bilinear_##flavor##_CnR(                                 \
1228     const arrtype* src, int step, CvSize ssize,                             \
1229     arrtype* dst, int dststep, CvSize dsize,                                \
1230     const double* matrix, int cn,                                           \
1231     const arrtype* fillval )                                                \
1232 {                                                                           \
1233     int x, y, k;                                                            \
1234     float A11 = (float)matrix[0], A12 = (float)matrix[1], A13 = (float)matrix[2];\
1235     float A21 = (float)matrix[3], A22 = (float)matrix[4], A23 = (float)matrix[5];\
1236     float A31 = (float)matrix[6], A32 = (float)matrix[7], A33 = (float)matrix[8];\
1237                                                                             \
1238     step /= sizeof(src[0]);                                                 \
1239     dststep /= sizeof(dst[0]);                                              \
1240                                                                             \
1241     for( y = 0; y < dsize.height; y++, dst += dststep )                     \
1242     {                                                                       \
1243         float xs0 = A12*y + A13;                                            \
1244         float ys0 = A22*y + A23;                                            \
1245         float ws = A32*y + A33;                                             \
1246                                                                             \
1247         for( x = 0; x < dsize.width; x++, xs0 += A11, ys0 += A21, ws += A31 )\
1248         {                                                                   \
1249             float inv_ws = 1.f/ws;                                          \
1250             float xs = xs0*inv_ws;                                          \
1251             float ys = ys0*inv_ws;                                          \
1252             int ixs = cvFloor(xs);                                          \
1253             int iys = cvFloor(ys);                                          \
1254             float a = xs - ixs;                                             \
1255             float b = ys - iys;                                             \
1256             float p0, p1;                                                   \
1257                                                                             \
1258             if( (unsigned)ixs < (unsigned)(ssize.width - 1) &&              \
1259                 (unsigned)iys < (unsigned)(ssize.height - 1) )              \
1260             {                                                               \
1261                 const arrtype* ptr = src + step*iys + ixs*cn;               \
1262                                                                             \
1263                 for( k = 0; k < cn; k++ )                                   \
1264                 {                                                           \
1265                     p0 = load_macro(ptr[k]) +                               \
1266                         a * (load_macro(ptr[k+cn]) - load_macro(ptr[k]));   \
1267                     p1 = load_macro(ptr[k+step]) +                          \
1268                         a * (load_macro(ptr[k+cn+step]) -                   \
1269                              load_macro(ptr[k+step]));                      \
1270                     dst[x*cn+k] = (arrtype)cast_macro(p0 + b*(p1 - p0));    \
1271                 }                                                           \
1272             }                                                               \
1273             else if( (unsigned)(ixs+1) < (unsigned)(ssize.width+1) &&       \
1274                      (unsigned)(iys+1) < (unsigned)(ssize.height+1))        \
1275             {                                                               \
1276                 int x0 = ICV_WARP_CLIP_X( ixs );                            \
1277                 int y0 = ICV_WARP_CLIP_Y( iys );                            \
1278                 int x1 = ICV_WARP_CLIP_X( ixs + 1 );                        \
1279                 int y1 = ICV_WARP_CLIP_Y( iys + 1 );                        \
1280                 const arrtype* ptr0, *ptr1, *ptr2, *ptr3;                   \
1281                                                                             \
1282                 ptr0 = src + y0*step + x0*cn;                               \
1283                 ptr1 = src + y0*step + x1*cn;                               \
1284                 ptr2 = src + y1*step + x0*cn;                               \
1285                 ptr3 = src + y1*step + x1*cn;                               \
1286                                                                             \
1287                 for( k = 0; k < cn; k++ )                                   \
1288                 {                                                           \
1289                     p0 = load_macro(ptr0[k]) +                              \
1290                         a * (load_macro(ptr1[k]) - load_macro(ptr0[k]));    \
1291                     p1 = load_macro(ptr2[k]) +                              \
1292                         a * (load_macro(ptr3[k]) - load_macro(ptr2[k]));    \
1293                     dst[x*cn+k] = (arrtype)cast_macro(p0 + b*(p1 - p0));    \
1294                 }                                                           \
1295             }                                                               \
1296             else if( fillval )                                              \
1297                 for( k = 0; k < cn; k++ )                                   \
1298                     dst[x*cn+k] = fillval[k];                               \
1299         }                                                                   \
1300     }                                                                       \
1301                                                                             \
1302     return CV_OK;                                                           \
1303 }
1304 
1305 
1306 #define ICV_WARP_SCALE_ALPHA(x) ((x)*(1./(ICV_WARP_MASK+1)))
1307 
1308 ICV_DEF_WARP_PERSPECTIVE_BILINEAR_FUNC( 8u, uchar, CV_8TO32F, cvRound )
1309 ICV_DEF_WARP_PERSPECTIVE_BILINEAR_FUNC( 16u, ushort, CV_NOP, cvRound )
1310 ICV_DEF_WARP_PERSPECTIVE_BILINEAR_FUNC( 32f, float, CV_NOP, CV_NOP )
1311 
1312 typedef CvStatus (CV_STDCALL * CvWarpPerspectiveFunc)(
1313     const void* src, int srcstep, CvSize ssize,
1314     void* dst, int dststep, CvSize dsize,
1315     const double* matrix, int cn, const void* fillval );
1316 
icvInitWarpPerspectiveTab(CvFuncTable * bilin_tab)1317 static void icvInitWarpPerspectiveTab( CvFuncTable* bilin_tab )
1318 {
1319     bilin_tab->fn_2d[CV_8U] = (void*)icvWarpPerspective_Bilinear_8u_CnR;
1320     bilin_tab->fn_2d[CV_16U] = (void*)icvWarpPerspective_Bilinear_16u_CnR;
1321     bilin_tab->fn_2d[CV_32F] = (void*)icvWarpPerspective_Bilinear_32f_CnR;
1322 }
1323 
1324 
1325 /////////////////////////// IPP warpperspective functions ////////////////////////////////
1326 
1327 icvWarpPerspectiveBack_8u_C1R_t icvWarpPerspectiveBack_8u_C1R_p = 0;
1328 icvWarpPerspectiveBack_8u_C3R_t icvWarpPerspectiveBack_8u_C3R_p = 0;
1329 icvWarpPerspectiveBack_8u_C4R_t icvWarpPerspectiveBack_8u_C4R_p = 0;
1330 icvWarpPerspectiveBack_32f_C1R_t icvWarpPerspectiveBack_32f_C1R_p = 0;
1331 icvWarpPerspectiveBack_32f_C3R_t icvWarpPerspectiveBack_32f_C3R_p = 0;
1332 icvWarpPerspectiveBack_32f_C4R_t icvWarpPerspectiveBack_32f_C4R_p = 0;
1333 
1334 icvWarpPerspective_8u_C1R_t icvWarpPerspective_8u_C1R_p = 0;
1335 icvWarpPerspective_8u_C3R_t icvWarpPerspective_8u_C3R_p = 0;
1336 icvWarpPerspective_8u_C4R_t icvWarpPerspective_8u_C4R_p = 0;
1337 icvWarpPerspective_32f_C1R_t icvWarpPerspective_32f_C1R_p = 0;
1338 icvWarpPerspective_32f_C3R_t icvWarpPerspective_32f_C3R_p = 0;
1339 icvWarpPerspective_32f_C4R_t icvWarpPerspective_32f_C4R_p = 0;
1340 
1341 typedef CvStatus (CV_STDCALL * CvWarpPerspectiveBackIPPFunc)
1342 ( const void* src, CvSize srcsize, int srcstep, CvRect srcroi,
1343   void* dst, int dststep, CvRect dstroi,
1344   const double* coeffs, int interpolation );
1345 
1346 //////////////////////////////////////////////////////////////////////////////////////////
1347 
1348 CV_IMPL void
cvWarpPerspective(const CvArr * srcarr,CvArr * dstarr,const CvMat * matrix,int flags,CvScalar fillval)1349 cvWarpPerspective( const CvArr* srcarr, CvArr* dstarr,
1350                    const CvMat* matrix, int flags, CvScalar fillval )
1351 {
1352     static CvFuncTable bilin_tab;
1353     static int inittab = 0;
1354 
1355     CV_FUNCNAME( "cvWarpPerspective" );
1356 
1357     __BEGIN__;
1358 
1359     CvMat srcstub, *src = (CvMat*)srcarr;
1360     CvMat dststub, *dst = (CvMat*)dstarr;
1361     int type, depth, cn;
1362     int method = flags & 3;
1363     double src_matrix[9], dst_matrix[9];
1364     double fillbuf[4];
1365     CvMat A = cvMat( 3, 3, CV_64F, src_matrix ),
1366           invA = cvMat( 3, 3, CV_64F, dst_matrix );
1367     CvWarpPerspectiveFunc func;
1368     CvSize ssize, dsize;
1369 
1370     if( method == CV_INTER_NN || method == CV_INTER_AREA )
1371         method = CV_INTER_LINEAR;
1372 
1373     if( !inittab )
1374     {
1375         icvInitWarpPerspectiveTab( &bilin_tab );
1376         inittab = 1;
1377     }
1378 
1379     CV_CALL( src = cvGetMat( srcarr, &srcstub ));
1380     CV_CALL( dst = cvGetMat( dstarr, &dststub ));
1381 
1382     if( !CV_ARE_TYPES_EQ( src, dst ))
1383         CV_ERROR( CV_StsUnmatchedFormats, "" );
1384 
1385     if( !CV_IS_MAT(matrix) || CV_MAT_CN(matrix->type) != 1 ||
1386         CV_MAT_DEPTH(matrix->type) < CV_32F || matrix->rows != 3 || matrix->cols != 3 )
1387         CV_ERROR( CV_StsBadArg,
1388         "Transformation matrix should be 3x3 floating-point single-channel matrix" );
1389 
1390     if( flags & CV_WARP_INVERSE_MAP )
1391         cvConvertScale( matrix, &invA );
1392     else
1393     {
1394         cvConvertScale( matrix, &A );
1395         cvInvert( &A, &invA, CV_SVD );
1396     }
1397 
1398     type = CV_MAT_TYPE(src->type);
1399     depth = CV_MAT_DEPTH(type);
1400     cn = CV_MAT_CN(type);
1401     if( cn > 4 )
1402         CV_ERROR( CV_BadNumChannels, "" );
1403 
1404     ssize = cvGetMatSize(src);
1405     dsize = cvGetMatSize(dst);
1406 
1407     if( icvWarpPerspectiveBack_8u_C1R_p )
1408     {
1409         CvWarpPerspectiveBackIPPFunc ipp_func =
1410             type == CV_8UC1 ? icvWarpPerspectiveBack_8u_C1R_p :
1411             type == CV_8UC3 ? icvWarpPerspectiveBack_8u_C3R_p :
1412             type == CV_8UC4 ? icvWarpPerspectiveBack_8u_C4R_p :
1413             type == CV_32FC1 ? icvWarpPerspectiveBack_32f_C1R_p :
1414             type == CV_32FC3 ? icvWarpPerspectiveBack_32f_C3R_p :
1415             type == CV_32FC4 ? icvWarpPerspectiveBack_32f_C4R_p : 0;
1416 
1417         if( ipp_func && CV_INTER_NN <= method && method <= CV_INTER_AREA &&
1418             MIN(ssize.width,ssize.height) >= 4 && MIN(dsize.width,dsize.height) >= 4 )
1419         {
1420             int srcstep = src->step ? src->step : CV_STUB_STEP;
1421             int dststep = dst->step ? dst->step : CV_STUB_STEP;
1422             CvStatus status;
1423             CvRect srcroi = {0, 0, ssize.width, ssize.height};
1424             CvRect dstroi = {0, 0, dsize.width, dsize.height};
1425 
1426             // this is not the most efficient way to fill outliers
1427             if( flags & CV_WARP_FILL_OUTLIERS )
1428                 cvSet( dst, fillval );
1429 
1430             status = ipp_func( src->data.ptr, ssize, srcstep, srcroi,
1431                                dst->data.ptr, dststep, dstroi,
1432                                invA.data.db, 1 << method );
1433             if( status >= 0 )
1434                 EXIT;
1435 
1436             ipp_func = type == CV_8UC1 ? icvWarpPerspective_8u_C1R_p :
1437                 type == CV_8UC3 ? icvWarpPerspective_8u_C3R_p :
1438                 type == CV_8UC4 ? icvWarpPerspective_8u_C4R_p :
1439                 type == CV_32FC1 ? icvWarpPerspective_32f_C1R_p :
1440                 type == CV_32FC3 ? icvWarpPerspective_32f_C3R_p :
1441                 type == CV_32FC4 ? icvWarpPerspective_32f_C4R_p : 0;
1442 
1443             if( ipp_func )
1444             {
1445                 if( flags & CV_WARP_INVERSE_MAP )
1446                     cvInvert( &invA, &A, CV_SVD );
1447 
1448                 status = ipp_func( src->data.ptr, ssize, srcstep, srcroi,
1449                                dst->data.ptr, dststep, dstroi,
1450                                A.data.db, 1 << method );
1451                 if( status >= 0 )
1452                     EXIT;
1453             }
1454         }
1455     }
1456 
1457     cvScalarToRawData( &fillval, fillbuf, CV_MAT_TYPE(src->type), 0 );
1458 
1459     /*if( method == CV_INTER_LINEAR )*/
1460     {
1461         func = (CvWarpPerspectiveFunc)bilin_tab.fn_2d[depth];
1462         if( !func )
1463             CV_ERROR( CV_StsUnsupportedFormat, "" );
1464 
1465         IPPI_CALL( func( src->data.ptr, src->step, ssize, dst->data.ptr,
1466                          dst->step, dsize, dst_matrix, cn,
1467                          flags & CV_WARP_FILL_OUTLIERS ? fillbuf : 0 ));
1468     }
1469 
1470     __END__;
1471 }
1472 
1473 
1474 /* Calculates coefficients of perspective transformation
1475  * which maps (xi,yi) to (ui,vi), (i=1,2,3,4):
1476  *
1477  *      c00*xi + c01*yi + c02
1478  * ui = ---------------------
1479  *      c20*xi + c21*yi + c22
1480  *
1481  *      c10*xi + c11*yi + c12
1482  * vi = ---------------------
1483  *      c20*xi + c21*yi + c22
1484  *
1485  * Coefficients are calculated by solving linear system:
1486  * / x0 y0  1  0  0  0 -x0*u0 -y0*u0 \ /c00\ /u0\
1487  * | x1 y1  1  0  0  0 -x1*u1 -y1*u1 | |c01| |u1|
1488  * | x2 y2  1  0  0  0 -x2*u2 -y2*u2 | |c02| |u2|
1489  * | x3 y3  1  0  0  0 -x3*u3 -y3*u3 |.|c10|=|u3|,
1490  * |  0  0  0 x0 y0  1 -x0*v0 -y0*v0 | |c11| |v0|
1491  * |  0  0  0 x1 y1  1 -x1*v1 -y1*v1 | |c12| |v1|
1492  * |  0  0  0 x2 y2  1 -x2*v2 -y2*v2 | |c20| |v2|
1493  * \  0  0  0 x3 y3  1 -x3*v3 -y3*v3 / \c21/ \v3/
1494  *
1495  * where:
1496  *   cij - matrix coefficients, c22 = 1
1497  */
1498 CV_IMPL CvMat*
cvGetPerspectiveTransform(const CvPoint2D32f * src,const CvPoint2D32f * dst,CvMat * matrix)1499 cvGetPerspectiveTransform( const CvPoint2D32f* src,
1500                           const CvPoint2D32f* dst,
1501                           CvMat* matrix )
1502 {
1503     CV_FUNCNAME( "cvGetPerspectiveTransform" );
1504 
1505     __BEGIN__;
1506 
1507     double a[8][8];
1508     double b[8], x[9];
1509 
1510     CvMat A = cvMat( 8, 8, CV_64FC1, a );
1511     CvMat B = cvMat( 8, 1, CV_64FC1, b );
1512     CvMat X = cvMat( 8, 1, CV_64FC1, x );
1513 
1514     int i;
1515 
1516     if( !src || !dst || !matrix )
1517         CV_ERROR( CV_StsNullPtr, "" );
1518 
1519     for( i = 0; i < 4; ++i )
1520     {
1521         a[i][0] = a[i+4][3] = src[i].x;
1522         a[i][1] = a[i+4][4] = src[i].y;
1523         a[i][2] = a[i+4][5] = 1;
1524         a[i][3] = a[i][4] = a[i][5] =
1525         a[i+4][0] = a[i+4][1] = a[i+4][2] = 0;
1526         a[i][6] = -src[i].x*dst[i].x;
1527         a[i][7] = -src[i].y*dst[i].x;
1528         a[i+4][6] = -src[i].x*dst[i].y;
1529         a[i+4][7] = -src[i].y*dst[i].y;
1530         b[i] = dst[i].x;
1531         b[i+4] = dst[i].y;
1532     }
1533 
1534     cvSolve( &A, &B, &X, CV_SVD );
1535     x[8] = 1;
1536 
1537     X = cvMat( 3, 3, CV_64FC1, x );
1538     cvConvert( &X, matrix );
1539 
1540     __END__;
1541 
1542     return matrix;
1543 }
1544 
1545 /* Calculates coefficients of affine transformation
1546  * which maps (xi,yi) to (ui,vi), (i=1,2,3):
1547  *
1548  * ui = c00*xi + c01*yi + c02
1549  *
1550  * vi = c10*xi + c11*yi + c12
1551  *
1552  * Coefficients are calculated by solving linear system:
1553  * / x0 y0  1  0  0  0 \ /c00\ /u0\
1554  * | x1 y1  1  0  0  0 | |c01| |u1|
1555  * | x2 y2  1  0  0  0 | |c02| |u2|
1556  * |  0  0  0 x0 y0  1 | |c10| |v0|
1557  * |  0  0  0 x1 y1  1 | |c11| |v1|
1558  * \  0  0  0 x2 y2  1 / |c12| |v2|
1559  *
1560  * where:
1561  *   cij - matrix coefficients
1562  */
1563 CV_IMPL CvMat*
cvGetAffineTransform(const CvPoint2D32f * src,const CvPoint2D32f * dst,CvMat * map_matrix)1564 cvGetAffineTransform( const CvPoint2D32f * src, const CvPoint2D32f * dst, CvMat * map_matrix )
1565 {
1566     CV_FUNCNAME( "cvGetAffineTransform" );
1567 
1568     __BEGIN__;
1569 
1570     CvMat mA, mX, mB;
1571     double A[6*6];
1572     double B[6];
1573 	double x[6];
1574     int i;
1575 
1576     cvInitMatHeader(&mA, 6, 6, CV_64F, A);
1577     cvInitMatHeader(&mB, 6, 1, CV_64F, B);
1578 	cvInitMatHeader(&mX, 6, 1, CV_64F, x);
1579 
1580 	if( !src || !dst || !map_matrix )
1581 		CV_ERROR( CV_StsNullPtr, "" );
1582 
1583     for( i = 0; i < 3; i++ )
1584     {
1585         int j = i*12;
1586         int k = i*12+6;
1587         A[j] = A[k+3] = src[i].x;
1588         A[j+1] = A[k+4] = src[i].y;
1589         A[j+2] = A[k+5] = 1;
1590         A[j+3] = A[j+4] = A[j+5] = 0;
1591         A[k] = A[k+1] = A[k+2] = 0;
1592         B[i*2] = dst[i].x;
1593         B[i*2+1] = dst[i].y;
1594     }
1595     cvSolve(&mA, &mB, &mX);
1596 
1597     mX = cvMat( 2, 3, CV_64FC1, x );
1598 	cvConvert( &mX, map_matrix );
1599 
1600 	__END__;
1601     return map_matrix;
1602 }
1603 
1604 /****************************************************************************************\
1605 *                          Generic Geometric Transformation: Remap                       *
1606 \****************************************************************************************/
1607 
1608 #define  ICV_DEF_REMAP_BILINEAR_FUNC( flavor, arrtype, load_macro, cast_macro ) \
1609 static CvStatus CV_STDCALL                                                      \
1610 icvRemap_Bilinear_##flavor##_CnR( const arrtype* src, int srcstep, CvSize ssize,\
1611                          arrtype* dst, int dststep, CvSize dsize,           \
1612                          const float* mapx, int mxstep,                     \
1613                          const float* mapy, int mystep,                     \
1614                          int cn, const arrtype* fillval )                   \
1615 {                                                                           \
1616     int i, j, k;                                                            \
1617     ssize.width--;                                                          \
1618     ssize.height--;                                                         \
1619                                                                             \
1620     srcstep /= sizeof(src[0]);                                              \
1621     dststep /= sizeof(dst[0]);                                              \
1622     mxstep /= sizeof(mapx[0]);                                              \
1623     mystep /= sizeof(mapy[0]);                                              \
1624                                                                             \
1625     for( i = 0; i < dsize.height; i++, dst += dststep,                      \
1626                                   mapx += mxstep, mapy += mystep )          \
1627     {                                                                       \
1628         for( j = 0; j < dsize.width; j++ )                                  \
1629         {                                                                   \
1630             float _x = mapx[j], _y = mapy[j];                               \
1631             int ix = cvFloor(_x), iy = cvFloor(_y);                         \
1632                                                                             \
1633             if( (unsigned)ix < (unsigned)ssize.width &&                     \
1634                 (unsigned)iy < (unsigned)ssize.height )                     \
1635             {                                                               \
1636                 const arrtype* s = src + iy*srcstep + ix*cn;                \
1637                 _x -= ix; _y -= iy;                                         \
1638                 for( k = 0; k < cn; k++, s++ )                              \
1639                 {                                                           \
1640                     float t0 = load_macro(s[0]), t1 = load_macro(s[srcstep]); \
1641                     t0 += _x*(load_macro(s[cn]) - t0);                      \
1642                     t1 += _x*(load_macro(s[srcstep + cn]) - t1);            \
1643                     dst[j*cn + k] = (arrtype)cast_macro(t0 + _y*(t1 - t0)); \
1644                 }                                                           \
1645             }                                                               \
1646             else if( fillval )                                              \
1647                 for( k = 0; k < cn; k++ )                                   \
1648                     dst[j*cn + k] = fillval[k];                             \
1649         }                                                                   \
1650     }                                                                       \
1651                                                                             \
1652     return CV_OK;                                                           \
1653 }
1654 
1655 
1656 #define  ICV_DEF_REMAP_BICUBIC_FUNC( flavor, arrtype, worktype,                 \
1657                                      load_macro, cast_macro1, cast_macro2 )     \
1658 static CvStatus CV_STDCALL                                                      \
1659 icvRemap_Bicubic_##flavor##_CnR( const arrtype* src, int srcstep, CvSize ssize, \
1660                          arrtype* dst, int dststep, CvSize dsize,               \
1661                          const float* mapx, int mxstep,                         \
1662                          const float* mapy, int mystep,                         \
1663                          int cn, const arrtype* fillval )                       \
1664 {                                                                               \
1665     int i, j, k;                                                                \
1666     ssize.width = MAX( ssize.width - 3, 0 );                                    \
1667     ssize.height = MAX( ssize.height - 3, 0 );                                  \
1668                                                                                 \
1669     srcstep /= sizeof(src[0]);                                                  \
1670     dststep /= sizeof(dst[0]);                                                  \
1671     mxstep /= sizeof(mapx[0]);                                                  \
1672     mystep /= sizeof(mapy[0]);                                                  \
1673                                                                                 \
1674     for( i = 0; i < dsize.height; i++, dst += dststep,                          \
1675                                   mapx += mxstep, mapy += mystep )              \
1676     {                                                                           \
1677         for( j = 0; j < dsize.width; j++ )                                      \
1678         {                                                                       \
1679             int ix = cvRound(mapx[j]*(1 << ICV_WARP_SHIFT));                    \
1680             int iy = cvRound(mapy[j]*(1 << ICV_WARP_SHIFT));                    \
1681             int ifx = ix & ICV_WARP_MASK;                                       \
1682             int ify = iy & ICV_WARP_MASK;                                       \
1683             ix >>= ICV_WARP_SHIFT;                                              \
1684             iy >>= ICV_WARP_SHIFT;                                              \
1685                                                                                 \
1686             if( (unsigned)(ix-1) < (unsigned)ssize.width &&                     \
1687                 (unsigned)(iy-1) < (unsigned)ssize.height )                     \
1688             {                                                                   \
1689                 for( k = 0; k < cn; k++ )                                       \
1690                 {                                                               \
1691                     const arrtype* s = src + (iy-1)*srcstep + ix*cn + k;        \
1692                                                                                 \
1693                     float t0 = load_macro(s[-cn])*icvCubicCoeffs[ifx*2 + 1] +   \
1694                                load_macro(s[0])*icvCubicCoeffs[ifx*2] +         \
1695                                load_macro(s[cn])*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2] +\
1696                                load_macro(s[cn*2])*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2+1];\
1697                                                                                 \
1698                     s += srcstep;                                               \
1699                                                                                 \
1700                     float t1 = load_macro(s[-cn])*icvCubicCoeffs[ifx*2 + 1] +   \
1701                                load_macro(s[0])*icvCubicCoeffs[ifx*2] +         \
1702                                load_macro(s[cn])*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2] +\
1703                                load_macro(s[cn*2])*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2+1];\
1704                                                                                 \
1705                     s += srcstep;                                               \
1706                                                                                 \
1707                     float t2 = load_macro(s[-cn])*icvCubicCoeffs[ifx*2 + 1] +   \
1708                                load_macro(s[0])*icvCubicCoeffs[ifx*2] +         \
1709                                load_macro(s[cn])*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2] +\
1710                                load_macro(s[cn*2])*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2+1];\
1711                                                                                 \
1712                     s += srcstep;                                               \
1713                                                                                 \
1714                     float t3 = load_macro(s[-cn])*icvCubicCoeffs[ifx*2 + 1] +   \
1715                                load_macro(s[0])*icvCubicCoeffs[ifx*2] +         \
1716                                load_macro(s[cn])*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2] +\
1717                                load_macro(s[cn*2])*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ifx)*2+1];\
1718                                                                                 \
1719                     worktype t = cast_macro1( t0*icvCubicCoeffs[ify*2 + 1] +    \
1720                                t1*icvCubicCoeffs[ify*2] +                       \
1721                                t2*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ify)*2] +  \
1722                                t3*icvCubicCoeffs[(ICV_CUBIC_TAB_SIZE-ify)*2+1] );\
1723                                                                                 \
1724                     dst[j*cn + k] = cast_macro2(t);                             \
1725                 }                                                               \
1726             }                                                                   \
1727             else if( fillval )                                                  \
1728                 for( k = 0; k < cn; k++ )                                       \
1729                     dst[j*cn + k] = fillval[k];                                 \
1730         }                                                                       \
1731     }                                                                           \
1732                                                                                 \
1733     return CV_OK;                                                               \
1734 }
1735 
1736 
1737 ICV_DEF_REMAP_BILINEAR_FUNC( 8u, uchar, CV_8TO32F, cvRound )
1738 ICV_DEF_REMAP_BILINEAR_FUNC( 16u, ushort, CV_NOP, cvRound )
1739 ICV_DEF_REMAP_BILINEAR_FUNC( 32f, float, CV_NOP, CV_NOP )
1740 
1741 ICV_DEF_REMAP_BICUBIC_FUNC( 8u, uchar, int, CV_8TO32F, cvRound, CV_FAST_CAST_8U )
1742 ICV_DEF_REMAP_BICUBIC_FUNC( 16u, ushort, int, CV_NOP, cvRound, CV_CAST_16U )
1743 ICV_DEF_REMAP_BICUBIC_FUNC( 32f, float, float, CV_NOP, CV_NOP, CV_NOP )
1744 
1745 typedef CvStatus (CV_STDCALL * CvRemapFunc)(
1746     const void* src, int srcstep, CvSize ssize,
1747     void* dst, int dststep, CvSize dsize,
1748     const float* mapx, int mxstep,
1749     const float* mapy, int mystep,
1750     int cn, const void* fillval );
1751 
icvInitRemapTab(CvFuncTable * bilinear_tab,CvFuncTable * bicubic_tab)1752 static void icvInitRemapTab( CvFuncTable* bilinear_tab, CvFuncTable* bicubic_tab )
1753 {
1754     bilinear_tab->fn_2d[CV_8U] = (void*)icvRemap_Bilinear_8u_CnR;
1755     bilinear_tab->fn_2d[CV_16U] = (void*)icvRemap_Bilinear_16u_CnR;
1756     bilinear_tab->fn_2d[CV_32F] = (void*)icvRemap_Bilinear_32f_CnR;
1757 
1758     bicubic_tab->fn_2d[CV_8U] = (void*)icvRemap_Bicubic_8u_CnR;
1759     bicubic_tab->fn_2d[CV_16U] = (void*)icvRemap_Bicubic_16u_CnR;
1760     bicubic_tab->fn_2d[CV_32F] = (void*)icvRemap_Bicubic_32f_CnR;
1761 }
1762 
1763 
1764 /******************** IPP remap functions *********************/
1765 
1766 typedef CvStatus (CV_STDCALL * CvRemapIPPFunc)(
1767     const void* src, CvSize srcsize, int srcstep, CvRect srcroi,
1768     const float* xmap, int xmapstep, const float* ymap, int ymapstep,
1769     void* dst, int dststep, CvSize dstsize, int interpolation );
1770 
1771 icvRemap_8u_C1R_t icvRemap_8u_C1R_p = 0;
1772 icvRemap_8u_C3R_t icvRemap_8u_C3R_p = 0;
1773 icvRemap_8u_C4R_t icvRemap_8u_C4R_p = 0;
1774 
1775 icvRemap_32f_C1R_t icvRemap_32f_C1R_p = 0;
1776 icvRemap_32f_C3R_t icvRemap_32f_C3R_p = 0;
1777 icvRemap_32f_C4R_t icvRemap_32f_C4R_p = 0;
1778 
1779 /**************************************************************/
1780 
1781 #define CV_REMAP_SHIFT 5
1782 #define CV_REMAP_MASK ((1 << CV_REMAP_SHIFT) - 1)
1783 
1784 #if CV_SSE2 && defined(__GNUC__)
1785 #define align(x) __attribute__ ((aligned (x)))
1786 #elif CV_SSE2 && (defined(__ICL) || defined _MSC_VER && _MSC_VER >= 1300)
1787 #define align(x) __declspec(align(x))
1788 #else
1789 #define align(x)
1790 #endif
1791 
icvRemapFixedPt_8u(const CvMat * src,CvMat * dst,const CvMat * xymap,const CvMat * amap,const uchar * fillval)1792 static void icvRemapFixedPt_8u( const CvMat* src, CvMat* dst,
1793     const CvMat* xymap, const CvMat* amap, const uchar* fillval )
1794 {
1795     const int TABSZ = 1 << (CV_REMAP_SHIFT*2);
1796     static ushort align(8) atab[TABSZ][4];
1797     static int inittab = 0;
1798 
1799     int x, y, cols = src->cols, rows = src->rows;
1800     const uchar* sptr0 = src->data.ptr;
1801     int sstep = src->step;
1802     uchar fv0 = fillval[0], fv1 = fillval[1], fv2 = fillval[2], fv3 = fillval[3];
1803     int cn = CV_MAT_CN(src->type);
1804 #if CV_SSE2
1805     const uchar* sptr1 = sptr0 + sstep;
1806     __m128i br = _mm_set1_epi32((cols-2) + ((rows-2)<<16));
1807     __m128i xy2ofs = _mm_set1_epi32(1 + (sstep << 16));
1808     __m128i z = _mm_setzero_si128();
1809     int align(16) iofs0[4], iofs1[4];
1810 #endif
1811 
1812     if( !inittab )
1813     {
1814         for( y = 0; y <= CV_REMAP_MASK; y++ )
1815             for( x = 0; x <= CV_REMAP_MASK; x++ )
1816             {
1817                 int k = (y << CV_REMAP_SHIFT) + x;
1818                 atab[k][0] = (ushort)((CV_REMAP_MASK+1 - y)*(CV_REMAP_MASK+1 - x));
1819                 atab[k][1] = (ushort)((CV_REMAP_MASK+1 - y)*x);
1820                 atab[k][2] = (ushort)(y*(CV_REMAP_MASK+1 - x));
1821                 atab[k][3] = (ushort)(y*x);
1822             }
1823         inittab = 1;
1824     }
1825 
1826     for( y = 0; y < rows; y++ )
1827     {
1828         const short* xy = (const short*)(xymap->data.ptr + xymap->step*y);
1829         const ushort* alpha = (const ushort*)(amap->data.ptr + amap->step*y);
1830         uchar* dptr = (uchar*)(dst->data.ptr + dst->step*y);
1831         int x = 0;
1832 
1833         if( cn == 1 )
1834         {
1835     #if CV_SSE2
1836             for( ; x <= cols - 8; x += 8 )
1837             {
1838                 __m128i xy0 = _mm_load_si128( (const __m128i*)(xy + x*2));
1839                 __m128i xy1 = _mm_load_si128( (const __m128i*)(xy + x*2 + 8));
1840                 // 0|0|0|0|... <= x0|y0|x1|y1|... < cols-1|rows-1|cols-1|rows-1|... ?
1841                 __m128i mask0 = _mm_cmpeq_epi32(_mm_or_si128(_mm_cmpgt_epi16(z, xy0),
1842                                                 _mm_cmpgt_epi16(xy0,br)), z);
1843                 __m128i mask1 = _mm_cmpeq_epi32(_mm_or_si128(_mm_cmpgt_epi16(z, xy1),
1844                                                 _mm_cmpgt_epi16(xy1,br)), z);
1845                 __m128i ofs0 = _mm_and_si128(_mm_madd_epi16( xy0, xy2ofs ), mask0 );
1846                 __m128i ofs1 = _mm_and_si128(_mm_madd_epi16( xy1, xy2ofs ), mask1 );
1847                 unsigned i0, i1;
1848                 __m128i v0, v1, v2, v3, a0, a1, b0, b1;
1849                 _mm_store_si128( (__m128i*)iofs0, ofs0 );
1850                 _mm_store_si128( (__m128i*)iofs1, ofs1 );
1851                 i0 = *(ushort*)(sptr0 + iofs0[0]) + (*(ushort*)(sptr0 + iofs0[1]) << 16);
1852                 i1 = *(ushort*)(sptr0 + iofs0[2]) + (*(ushort*)(sptr0 + iofs0[3]) << 16);
1853                 v0 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(i0), _mm_cvtsi32_si128(i1));
1854                 i0 = *(ushort*)(sptr1 + iofs0[0]) + (*(ushort*)(sptr1 + iofs0[1]) << 16);
1855                 i1 = *(ushort*)(sptr1 + iofs0[2]) + (*(ushort*)(sptr1 + iofs0[3]) << 16);
1856                 v1 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(i0), _mm_cvtsi32_si128(i1));
1857                 v0 = _mm_unpacklo_epi8(v0, z);
1858                 v1 = _mm_unpacklo_epi8(v1, z);
1859 
1860                 a0 = _mm_unpacklo_epi32(_mm_loadl_epi64((__m128i*)atab[alpha[x]]),
1861                                         _mm_loadl_epi64((__m128i*)atab[alpha[x+1]]));
1862                 a1 = _mm_unpacklo_epi32(_mm_loadl_epi64((__m128i*)atab[alpha[x+2]]),
1863                                         _mm_loadl_epi64((__m128i*)atab[alpha[x+3]]));
1864                 b0 = _mm_unpacklo_epi64(a0, a1);
1865                 b1 = _mm_unpackhi_epi64(a0, a1);
1866                 v0 = _mm_madd_epi16(v0, b0);
1867                 v1 = _mm_madd_epi16(v1, b1);
1868                 v0 = _mm_and_si128(_mm_add_epi32(v0, v1), mask0);
1869 
1870                 i0 = *(ushort*)(sptr0 + iofs1[0]) + (*(ushort*)(sptr0 + iofs1[1]) << 16);
1871                 i1 = *(ushort*)(sptr0 + iofs1[2]) + (*(ushort*)(sptr0 + iofs1[3]) << 16);
1872                 v2 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(i0), _mm_cvtsi32_si128(i1));
1873                 i0 = *(ushort*)(sptr1 + iofs1[0]) + (*(ushort*)(sptr1 + iofs1[1]) << 16);
1874                 i1 = *(ushort*)(sptr1 + iofs1[2]) + (*(ushort*)(sptr1 + iofs1[3]) << 16);
1875                 v3 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(i0), _mm_cvtsi32_si128(i1));
1876                 v2 = _mm_unpacklo_epi8(v2, z);
1877                 v3 = _mm_unpacklo_epi8(v3, z);
1878 
1879                 a0 = _mm_unpacklo_epi32(_mm_loadl_epi64((__m128i*)atab[alpha[x+4]]),
1880                                         _mm_loadl_epi64((__m128i*)atab[alpha[x+5]]));
1881                 a1 = _mm_unpacklo_epi32(_mm_loadl_epi64((__m128i*)atab[alpha[x+6]]),
1882                                         _mm_loadl_epi64((__m128i*)atab[alpha[x+7]]));
1883                 b0 = _mm_unpacklo_epi64(a0, a1);
1884                 b1 = _mm_unpackhi_epi64(a0, a1);
1885                 v2 = _mm_madd_epi16(v2, b0);
1886                 v3 = _mm_madd_epi16(v3, b1);
1887                 v2 = _mm_and_si128(_mm_add_epi32(v2, v3), mask1);
1888 
1889                 v0 = _mm_srai_epi32(v0, CV_REMAP_SHIFT*2);
1890                 v2 = _mm_srai_epi32(v2, CV_REMAP_SHIFT*2);
1891                 v0 = _mm_packus_epi16(_mm_packs_epi32(v0, v2), z);
1892                 _mm_storel_epi64( (__m128i*)(dptr + x), v0 );
1893             }
1894     #endif
1895 
1896             for( ; x < cols; x++ )
1897             {
1898                 int xi = xy[x*2], yi = xy[x*2+1];
1899                 if( (unsigned)yi >= (unsigned)(rows - 1) ||
1900                     (unsigned)xi >= (unsigned)(cols - 1))
1901                 {
1902                     dptr[x] = fv0;
1903                 }
1904                 else
1905                 {
1906                     const uchar* sptr = sptr0 + sstep*yi + xi;
1907                     const ushort* a = atab[alpha[x]];
1908                     dptr[x] = (uchar)((sptr[0]*a[0] + sptr[1]*a[1] + sptr[sstep]*a[2] +
1909                                       sptr[sstep+1]*a[3])>>CV_REMAP_SHIFT*2);
1910                 }
1911             }
1912         }
1913         else if( cn == 3 )
1914         {
1915             for( ; x < cols; x++ )
1916             {
1917                 int xi = xy[x*2], yi = xy[x*2+1];
1918                 if( (unsigned)yi >= (unsigned)(rows - 1) ||
1919                     (unsigned)xi >= (unsigned)(cols - 1))
1920                 {
1921                     dptr[x*3] = fv0; dptr[x*3+1] = fv1; dptr[x*3+2] = fv2;
1922                 }
1923                 else
1924                 {
1925                     const uchar* sptr = sptr0 + sstep*yi + xi*3;
1926                     const ushort* a = atab[alpha[x]];
1927                     int v0, v1, v2;
1928                     v0 = (sptr[0]*a[0] + sptr[3]*a[1] +
1929                         sptr[sstep]*a[2] + sptr[sstep+3]*a[3])>>CV_REMAP_SHIFT*2;
1930                     v1 = (sptr[1]*a[0] + sptr[4]*a[1] +
1931                         sptr[sstep+1]*a[2] + sptr[sstep+4]*a[3])>>CV_REMAP_SHIFT*2;
1932                     v2 = (sptr[2]*a[0] + sptr[5]*a[1] +
1933                         sptr[sstep+2]*a[2] + sptr[sstep+5]*a[3])>>CV_REMAP_SHIFT*2;
1934                     dptr[x*3] = (uchar)v0; dptr[x*3+1] = (uchar)v1; dptr[x*3+2] = (uchar)v2;
1935                 }
1936             }
1937         }
1938         else
1939         {
1940             assert( cn == 4 );
1941             for( ; x < cols; x++ )
1942             {
1943                 int xi = xy[x*2], yi = xy[x*2+1];
1944                 if( (unsigned)yi >= (unsigned)(rows - 1) ||
1945                     (unsigned)xi >= (unsigned)(cols - 1))
1946                 {
1947                     dptr[x*4] = fv0; dptr[x*4+1] = fv1;
1948                     dptr[x*4+2] = fv2; dptr[x*4+3] = fv3;
1949                 }
1950                 else
1951                 {
1952                     const uchar* sptr = sptr0 + sstep*yi + xi*3;
1953                     const ushort* a = atab[alpha[x]];
1954                     int v0, v1;
1955                     v0 = (sptr[0]*a[0] + sptr[4]*a[1] +
1956                         sptr[sstep]*a[2] + sptr[sstep+3]*a[3])>>CV_REMAP_SHIFT*2;
1957                     v1 = (sptr[1]*a[0] + sptr[5]*a[1] +
1958                         sptr[sstep+1]*a[2] + sptr[sstep+5]*a[3])>>CV_REMAP_SHIFT*2;
1959                     dptr[x*4] = (uchar)v0; dptr[x*4+1] = (uchar)v1;
1960                     v0 = (sptr[2]*a[0] + sptr[6]*a[1] +
1961                         sptr[sstep+2]*a[2] + sptr[sstep+6]*a[3])>>CV_REMAP_SHIFT*2;
1962                     v1 = (sptr[3]*a[0] + sptr[7]*a[1] +
1963                         sptr[sstep+3]*a[2] + sptr[sstep+7]*a[3])>>CV_REMAP_SHIFT*2;
1964                     dptr[x*4+2] = (uchar)v0; dptr[x*4+3] = (uchar)v1;
1965                 }
1966             }
1967         }
1968     }
1969 }
1970 
1971 
1972 CV_IMPL void
cvRemap(const CvArr * srcarr,CvArr * dstarr,const CvArr * _mapx,const CvArr * _mapy,int flags,CvScalar fillval)1973 cvRemap( const CvArr* srcarr, CvArr* dstarr,
1974          const CvArr* _mapx, const CvArr* _mapy,
1975          int flags, CvScalar fillval )
1976 {
1977     static CvFuncTable bilinear_tab;
1978     static CvFuncTable bicubic_tab;
1979     static int inittab = 0;
1980 
1981     CV_FUNCNAME( "cvRemap" );
1982 
1983     __BEGIN__;
1984 
1985     CvMat srcstub, *src = (CvMat*)srcarr;
1986     CvMat dststub, *dst = (CvMat*)dstarr;
1987     CvMat mxstub, *mapx = (CvMat*)_mapx;
1988     CvMat mystub, *mapy = (CvMat*)_mapy;
1989     int type, depth, cn;
1990     bool fltremap;
1991     int method = flags & 3;
1992     double fillbuf[4];
1993     CvSize ssize, dsize;
1994 
1995     if( !inittab )
1996     {
1997         icvInitRemapTab( &bilinear_tab, &bicubic_tab );
1998         icvInitLinearCoeffTab();
1999         icvInitCubicCoeffTab();
2000         inittab = 1;
2001     }
2002 
2003     CV_CALL( src = cvGetMat( srcarr, &srcstub ));
2004     CV_CALL( dst = cvGetMat( dstarr, &dststub ));
2005     CV_CALL( mapx = cvGetMat( mapx, &mxstub ));
2006     CV_CALL( mapy = cvGetMat( mapy, &mystub ));
2007 
2008     if( !CV_ARE_TYPES_EQ( src, dst ))
2009         CV_ERROR( CV_StsUnmatchedFormats, "" );
2010 
2011     if( CV_MAT_TYPE(mapx->type) == CV_16SC1 && CV_MAT_TYPE(mapy->type) == CV_16SC2 )
2012     {
2013         CvMat* temp;
2014         CV_SWAP(mapx, mapy, temp);
2015     }
2016 
2017     if( (CV_MAT_TYPE(mapx->type) != CV_32FC1 || CV_MAT_TYPE(mapy->type) != CV_32FC1) &&
2018         (CV_MAT_TYPE(mapx->type) != CV_16SC2 || CV_MAT_TYPE(mapy->type) != CV_16SC1))
2019         CV_ERROR( CV_StsUnmatchedFormats, "Either both map arrays must have 32fC1 type, "
2020         "or one of them must be 16sC2 and the other one must be 16sC1" );
2021 
2022     if( !CV_ARE_SIZES_EQ( mapx, mapy ) || !CV_ARE_SIZES_EQ( mapx, dst ))
2023         CV_ERROR( CV_StsUnmatchedSizes,
2024         "Both map arrays and the destination array must have the same size" );
2025 
2026     fltremap = CV_MAT_TYPE(mapx->type) == CV_32FC1;
2027     type = CV_MAT_TYPE(src->type);
2028     depth = CV_MAT_DEPTH(type);
2029     cn = CV_MAT_CN(type);
2030     if( cn > 4 )
2031         CV_ERROR( CV_BadNumChannels, "" );
2032 
2033     ssize = cvGetMatSize(src);
2034     dsize = cvGetMatSize(dst);
2035 
2036     cvScalarToRawData( &fillval, fillbuf, CV_MAT_TYPE(src->type), 0 );
2037 
2038     if( !fltremap )
2039     {
2040         if( CV_MAT_TYPE(src->type) != CV_8UC1 && CV_MAT_TYPE(src->type) != CV_8UC3 &&
2041             CV_MAT_TYPE(src->type) != CV_8UC4 )
2042             CV_ERROR( CV_StsUnsupportedFormat,
2043             "Only 8-bit input/output is supported by the fixed-point variant of cvRemap" );
2044         icvRemapFixedPt_8u( src, dst, mapx, mapy, (uchar*)fillbuf );
2045         EXIT;
2046     }
2047 
2048     if( icvRemap_8u_C1R_p )
2049     {
2050         CvRemapIPPFunc ipp_func =
2051             type == CV_8UC1 ? icvRemap_8u_C1R_p :
2052             type == CV_8UC3 ? icvRemap_8u_C3R_p :
2053             type == CV_8UC4 ? icvRemap_8u_C4R_p :
2054             type == CV_32FC1 ? icvRemap_32f_C1R_p :
2055             type == CV_32FC3 ? icvRemap_32f_C3R_p :
2056             type == CV_32FC4 ? icvRemap_32f_C4R_p : 0;
2057 
2058         if( ipp_func )
2059         {
2060             int srcstep = src->step ? src->step : CV_STUB_STEP;
2061             int dststep = dst->step ? dst->step : CV_STUB_STEP;
2062             int mxstep = mapx->step ? mapx->step : CV_STUB_STEP;
2063             int mystep = mapy->step ? mapy->step : CV_STUB_STEP;
2064             CvStatus status;
2065             CvRect srcroi = {0, 0, ssize.width, ssize.height};
2066 
2067             // this is not the most efficient way to fill outliers
2068             if( flags & CV_WARP_FILL_OUTLIERS )
2069                 cvSet( dst, fillval );
2070 
2071             status = ipp_func( src->data.ptr, ssize, srcstep, srcroi,
2072                                mapx->data.fl, mxstep, mapy->data.fl, mystep,
2073                                dst->data.ptr, dststep, dsize,
2074                                1 << (method == CV_INTER_NN || method == CV_INTER_LINEAR ||
2075                                method == CV_INTER_CUBIC ? method : CV_INTER_LINEAR) );
2076             if( status >= 0 )
2077                 EXIT;
2078         }
2079     }
2080 
2081     {
2082         CvRemapFunc func = method == CV_INTER_CUBIC ?
2083             (CvRemapFunc)bicubic_tab.fn_2d[depth] :
2084             (CvRemapFunc)bilinear_tab.fn_2d[depth];
2085 
2086         if( !func )
2087             CV_ERROR( CV_StsUnsupportedFormat, "" );
2088 
2089         func( src->data.ptr, src->step, ssize, dst->data.ptr, dst->step, dsize,
2090               mapx->data.fl, mapx->step, mapy->data.fl, mapy->step,
2091               cn, flags & CV_WARP_FILL_OUTLIERS ? fillbuf : 0 );
2092     }
2093 
2094     __END__;
2095 }
2096 
2097 CV_IMPL void
cvConvertMaps(const CvArr * arrx,const CvArr * arry,CvArr * arrxy,CvArr * arra)2098 cvConvertMaps( const CvArr* arrx, const CvArr* arry,
2099                CvArr* arrxy, CvArr* arra )
2100 {
2101     CV_FUNCNAME( "cvConvertMaps" );
2102 
2103     __BEGIN__;
2104 
2105     CvMat xstub, *mapx = cvGetMat( arrx, &xstub );
2106     CvMat ystub, *mapy = cvGetMat( arry, &ystub );
2107     CvMat xystub, *mapxy = cvGetMat( arrxy, &xystub );
2108     CvMat astub, *mapa = cvGetMat( arra, &astub );
2109     int x, y, cols = mapx->cols, rows = mapx->rows;
2110 
2111     CV_ASSERT( CV_ARE_SIZES_EQ(mapx, mapy) && CV_ARE_TYPES_EQ(mapx, mapy) &&
2112         CV_MAT_TYPE(mapx->type) == CV_32FC1 &&
2113         CV_ARE_SIZES_EQ(mapxy, mapx) && CV_ARE_SIZES_EQ(mapxy, mapa) &&
2114         CV_MAT_TYPE(mapxy->type) == CV_16SC2 &&
2115         CV_MAT_TYPE(mapa->type) == CV_16SC1 );
2116 
2117     for( y = 0; y < rows; y++ )
2118     {
2119         const float* xrow = (const float*)(mapx->data.ptr + mapx->step*y);
2120         const float* yrow = (const float*)(mapy->data.ptr + mapy->step*y);
2121         short* xy = (short*)(mapxy->data.ptr + mapxy->step*y);
2122         short* alpha = (short*)(mapa->data.ptr + mapa->step*y);
2123 
2124         for( x = 0; x < cols; x++ )
2125         {
2126             int xi = cvRound(xrow[x]*(1 << CV_REMAP_SHIFT));
2127             int yi = cvRound(yrow[x]*(1 << CV_REMAP_SHIFT));
2128             xy[x*2] = (short)(xi >> CV_REMAP_SHIFT);
2129             xy[x*2+1] = (short)(yi >> CV_REMAP_SHIFT);
2130             alpha[x] = (short)((xi & CV_REMAP_MASK) + ((yi & CV_REMAP_MASK)<<CV_REMAP_SHIFT));
2131         }
2132     }
2133 
2134     __END__;
2135 }
2136 
2137 
2138 /****************************************************************************************\
2139 *                                   Log-Polar Transform                                  *
2140 \****************************************************************************************/
2141 
2142 /* now it is done via Remap; more correct implementation should use
2143    some super-sampling technique outside of the "fovea" circle */
2144 CV_IMPL void
cvLogPolar(const CvArr * srcarr,CvArr * dstarr,CvPoint2D32f center,double M,int flags)2145 cvLogPolar( const CvArr* srcarr, CvArr* dstarr,
2146             CvPoint2D32f center, double M, int flags )
2147 {
2148     CvMat* mapx = 0;
2149     CvMat* mapy = 0;
2150     double* exp_tab = 0;
2151     float* buf = 0;
2152 
2153     CV_FUNCNAME( "cvLogPolar" );
2154 
2155     __BEGIN__;
2156 
2157     CvMat srcstub, *src = (CvMat*)srcarr;
2158     CvMat dststub, *dst = (CvMat*)dstarr;
2159     CvSize ssize, dsize;
2160 
2161     CV_CALL( src = cvGetMat( srcarr, &srcstub ));
2162     CV_CALL( dst = cvGetMat( dstarr, &dststub ));
2163 
2164     if( !CV_ARE_TYPES_EQ( src, dst ))
2165         CV_ERROR( CV_StsUnmatchedFormats, "" );
2166 
2167     if( M <= 0 )
2168         CV_ERROR( CV_StsOutOfRange, "M should be >0" );
2169 
2170     ssize = cvGetMatSize(src);
2171     dsize = cvGetMatSize(dst);
2172 
2173     CV_CALL( mapx = cvCreateMat( dsize.height, dsize.width, CV_32F ));
2174     CV_CALL( mapy = cvCreateMat( dsize.height, dsize.width, CV_32F ));
2175 
2176     if( !(flags & CV_WARP_INVERSE_MAP) )
2177     {
2178         int phi, rho;
2179 
2180         CV_CALL( exp_tab = (double*)cvAlloc( dsize.width*sizeof(exp_tab[0])) );
2181 
2182         for( rho = 0; rho < dst->width; rho++ )
2183             exp_tab[rho] = exp(rho/M);
2184 
2185         for( phi = 0; phi < dsize.height; phi++ )
2186         {
2187             double cp = cos(phi*2*CV_PI/dsize.height);
2188             double sp = sin(phi*2*CV_PI/dsize.height);
2189             float* mx = (float*)(mapx->data.ptr + phi*mapx->step);
2190             float* my = (float*)(mapy->data.ptr + phi*mapy->step);
2191 
2192             for( rho = 0; rho < dsize.width; rho++ )
2193             {
2194                 double r = exp_tab[rho];
2195                 double x = r*cp + center.x;
2196                 double y = r*sp + center.y;
2197 
2198                 mx[rho] = (float)x;
2199                 my[rho] = (float)y;
2200             }
2201         }
2202     }
2203     else
2204     {
2205         int x, y;
2206         CvMat bufx, bufy, bufp, bufa;
2207         double ascale = (ssize.width-1)/(2*CV_PI);
2208 
2209         CV_CALL( buf = (float*)cvAlloc( 4*dsize.width*sizeof(buf[0]) ));
2210 
2211         bufx = cvMat( 1, dsize.width, CV_32F, buf );
2212         bufy = cvMat( 1, dsize.width, CV_32F, buf + dsize.width );
2213         bufp = cvMat( 1, dsize.width, CV_32F, buf + dsize.width*2 );
2214         bufa = cvMat( 1, dsize.width, CV_32F, buf + dsize.width*3 );
2215 
2216         for( x = 0; x < dsize.width; x++ )
2217             bufx.data.fl[x] = (float)x - center.x;
2218 
2219         for( y = 0; y < dsize.height; y++ )
2220         {
2221             float* mx = (float*)(mapx->data.ptr + y*mapx->step);
2222             float* my = (float*)(mapy->data.ptr + y*mapy->step);
2223 
2224             for( x = 0; x < dsize.width; x++ )
2225                 bufy.data.fl[x] = (float)y - center.y;
2226 
2227 #if 1
2228             cvCartToPolar( &bufx, &bufy, &bufp, &bufa );
2229 
2230             for( x = 0; x < dsize.width; x++ )
2231                 bufp.data.fl[x] += 1.f;
2232 
2233             cvLog( &bufp, &bufp );
2234 
2235             for( x = 0; x < dsize.width; x++ )
2236             {
2237                 double rho = bufp.data.fl[x]*M;
2238                 double phi = bufa.data.fl[x]*ascale;
2239 
2240                 mx[x] = (float)rho;
2241                 my[x] = (float)phi;
2242             }
2243 #else
2244             for( x = 0; x < dsize.width; x++ )
2245             {
2246                 double xx = bufx.data.fl[x];
2247                 double yy = bufy.data.fl[x];
2248 
2249                 double p = log(sqrt(xx*xx + yy*yy) + 1.)*M;
2250                 double a = atan2(yy,xx);
2251                 if( a < 0 )
2252                     a = 2*CV_PI + a;
2253                 a *= ascale;
2254 
2255                 mx[x] = (float)p;
2256                 my[x] = (float)a;
2257             }
2258 #endif
2259         }
2260     }
2261 
2262     cvRemap( src, dst, mapx, mapy, flags, cvScalarAll(0) );
2263 
2264     __END__;
2265 
2266     cvFree( &exp_tab );
2267     cvFree( &buf );
2268     cvReleaseMat( &mapx );
2269     cvReleaseMat( &mapy );
2270 }
2271 
2272 /* End of file. */
2273