• 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 #include "_cxcore.h"
43 
44 /****************************************************************************************\
45 *                           [scaled] Identity matrix initialization                      *
46 \****************************************************************************************/
47 
48 CV_IMPL void
cvSetIdentity(CvArr * array,CvScalar value)49 cvSetIdentity( CvArr* array, CvScalar value )
50 {
51     CV_FUNCNAME( "cvSetIdentity" );
52 
53     __BEGIN__;
54 
55     CvMat stub, *mat = (CvMat*)array;
56     CvSize size;
57     int i, k, len, step;
58     int type, pix_size;
59     uchar* data = 0;
60     double buf[4];
61 
62     if( !CV_IS_MAT( mat ))
63     {
64         int coi = 0;
65         CV_CALL( mat = cvGetMat( mat, &stub, &coi ));
66         if( coi != 0 )
67             CV_ERROR( CV_BadCOI, "coi is not supported" );
68     }
69 
70     size = cvGetMatSize( mat );
71     len = CV_IMIN( size.width, size.height );
72 
73     type = CV_MAT_TYPE(mat->type);
74     pix_size = CV_ELEM_SIZE(type);
75     size.width *= pix_size;
76 
77     if( CV_IS_MAT_CONT( mat->type ))
78     {
79         size.width *= size.height;
80         size.height = 1;
81     }
82 
83     data = mat->data.ptr;
84     step = mat->step;
85     if( step == 0 )
86         step = CV_STUB_STEP;
87     IPPI_CALL( icvSetZero_8u_C1R( data, step, size ));
88     step += pix_size;
89 
90     if( type == CV_32FC1 )
91     {
92         float val = (float)value.val[0];
93         float* _data = (float*)data;
94         step /= sizeof(_data[0]);
95         len *= step;
96 
97         for( i = 0; i < len; i += step )
98             _data[i] = val;
99     }
100     else if( type == CV_64FC1 )
101     {
102         double val = value.val[0];
103         double* _data = (double*)data;
104         step /= sizeof(_data[0]);
105         len *= step;
106 
107         for( i = 0; i < len; i += step )
108             _data[i] = val;
109     }
110     else
111     {
112         uchar* val_ptr = (uchar*)buf;
113         cvScalarToRawData( &value, buf, type, 0 );
114         len *= step;
115 
116         for( i = 0; i < len; i += step )
117             for( k = 0; k < pix_size; k++ )
118                 data[i+k] = val_ptr[k];
119     }
120 
121     __END__;
122 }
123 
124 
125 /****************************************************************************************\
126 *                                    Trace of the matrix                                 *
127 \****************************************************************************************/
128 
129 CV_IMPL CvScalar
cvTrace(const CvArr * array)130 cvTrace( const CvArr* array )
131 {
132     CvScalar sum = {{0,0,0,0}};
133 
134     CV_FUNCNAME( "cvTrace" );
135 
136     __BEGIN__;
137 
138     CvMat stub, *mat = 0;
139 
140     if( CV_IS_MAT( array ))
141     {
142         mat = (CvMat*)array;
143         int type = CV_MAT_TYPE(mat->type);
144         int size = MIN(mat->rows,mat->cols);
145         uchar* data = mat->data.ptr;
146 
147         if( type == CV_32FC1 )
148         {
149             int step = mat->step + sizeof(float);
150 
151             for( ; size--; data += step )
152                 sum.val[0] += *(float*)data;
153             EXIT;
154         }
155 
156         if( type == CV_64FC1 )
157         {
158             int step = mat->step + sizeof(double);
159 
160             for( ; size--; data += step )
161                 sum.val[0] += *(double*)data;
162             EXIT;
163         }
164     }
165 
166     CV_CALL( mat = cvGetDiag( array, &stub ));
167     CV_CALL( sum = cvSum( mat ));
168 
169     __END__;
170 
171     return sum;
172 }
173 
174 
175 /****************************************************************************************\
176 *                                     Matrix transpose                                   *
177 \****************************************************************************************/
178 
179 /////////////////// macros for inplace transposition of square matrix ////////////////////
180 
181 #define ICV_DEF_TRANSP_INP_CASE_C1( \
182     arrtype, len )                  \
183 {                                   \
184     arrtype* arr1 = arr;            \
185     step /= sizeof(arr[0]);         \
186                                     \
187     while( --len )                  \
188     {                               \
189         arr += step, arr1++;        \
190         arrtype* arr2 = arr;        \
191         arrtype* arr3 = arr1;       \
192                                     \
193         do                          \
194         {                           \
195             arrtype t0 = arr2[0];   \
196             arrtype t1 = arr3[0];   \
197             arr2[0] = t1;           \
198             arr3[0] = t0;           \
199                                     \
200             arr2++;                 \
201             arr3 += step;           \
202         }                           \
203         while( arr2 != arr3  );     \
204     }                               \
205 }
206 
207 
208 #define ICV_DEF_TRANSP_INP_CASE_C3( \
209     arrtype, len )                  \
210 {                                   \
211     arrtype* arr1 = arr;            \
212     int y;                          \
213     step /= sizeof(arr[0]);         \
214                                     \
215     for( y = 1; y < len; y++ )      \
216     {                               \
217         arr += step, arr1 += 3;     \
218         arrtype* arr2 = arr;        \
219         arrtype* arr3 = arr1;       \
220                                     \
221         for( ; arr2!=arr3; arr2+=3, \
222                         arr3+=step )\
223         {                           \
224             arrtype t0 = arr2[0];   \
225             arrtype t1 = arr3[0];   \
226             arr2[0] = t1;           \
227             arr3[0] = t0;           \
228             t0 = arr2[1];           \
229             t1 = arr3[1];           \
230             arr2[1] = t1;           \
231             arr3[1] = t0;           \
232             t0 = arr2[2];           \
233             t1 = arr3[2];           \
234             arr2[2] = t1;           \
235             arr3[2] = t0;           \
236         }                           \
237     }                               \
238 }
239 
240 
241 #define ICV_DEF_TRANSP_INP_CASE_C4( \
242     arrtype, len )                  \
243 {                                   \
244     arrtype* arr1 = arr;            \
245     int y;                          \
246     step /= sizeof(arr[0]);         \
247                                     \
248     for( y = 1; y < len; y++ )      \
249     {                               \
250         arr += step, arr1 += 4;     \
251         arrtype* arr2 = arr;        \
252         arrtype* arr3 = arr1;       \
253                                     \
254         for( ; arr2!=arr3; arr2+=4, \
255                         arr3+=step )\
256         {                           \
257             arrtype t0 = arr2[0];   \
258             arrtype t1 = arr3[0];   \
259             arr2[0] = t1;           \
260             arr3[0] = t0;           \
261             t0 = arr2[1];           \
262             t1 = arr3[1];           \
263             arr2[1] = t1;           \
264             arr3[1] = t0;           \
265             t0 = arr2[2];           \
266             t1 = arr3[2];           \
267             arr2[2] = t1;           \
268             arr3[2] = t0;           \
269             t0 = arr2[3];           \
270             t1 = arr3[3];           \
271             arr2[3] = t1;           \
272             arr3[3] = t0;           \
273         }                           \
274     }                               \
275 }
276 
277 
278 //////////////// macros for non-inplace transposition of rectangular matrix //////////////
279 
280 #define ICV_DEF_TRANSP_CASE_C1( arrtype )       \
281 {                                               \
282     int x, y;                                   \
283     srcstep /= sizeof(src[0]);                  \
284     dststep /= sizeof(dst[0]);                  \
285                                                 \
286     for( y = 0; y <= size.height - 2; y += 2,   \
287                 src += 2*srcstep, dst += 2 )    \
288     {                                           \
289         const arrtype* src1 = src + srcstep;    \
290         arrtype* dst1 = dst;                    \
291                                                 \
292         for( x = 0; x <= size.width - 2;        \
293                 x += 2, dst1 += dststep )       \
294         {                                       \
295             arrtype t0 = src[x];                \
296             arrtype t1 = src1[x];               \
297             dst1[0] = t0;                       \
298             dst1[1] = t1;                       \
299             dst1 += dststep;                    \
300                                                 \
301             t0 = src[x + 1];                    \
302             t1 = src1[x + 1];                   \
303             dst1[0] = t0;                       \
304             dst1[1] = t1;                       \
305         }                                       \
306                                                 \
307         if( x < size.width )                    \
308         {                                       \
309             arrtype t0 = src[x];                \
310             arrtype t1 = src1[x];               \
311             dst1[0] = t0;                       \
312             dst1[1] = t1;                       \
313         }                                       \
314     }                                           \
315                                                 \
316     if( y < size.height )                       \
317     {                                           \
318         arrtype* dst1 = dst;                    \
319         for( x = 0; x <= size.width - 2;        \
320                 x += 2, dst1 += 2*dststep )     \
321         {                                       \
322             arrtype t0 = src[x];                \
323             arrtype t1 = src[x + 1];            \
324             dst1[0] = t0;                       \
325             dst1[dststep] = t1;                 \
326         }                                       \
327                                                 \
328         if( x < size.width )                    \
329         {                                       \
330             arrtype t0 = src[x];                \
331             dst1[0] = t0;                       \
332         }                                       \
333     }                                           \
334 }
335 
336 
337 #define ICV_DEF_TRANSP_CASE_C3( arrtype )       \
338 {                                               \
339     size.width *= 3;                            \
340     srcstep /= sizeof(src[0]);                  \
341     dststep /= sizeof(dst[0]);                  \
342                                                 \
343     for( ; size.height--; src+=srcstep, dst+=3 )\
344     {                                           \
345         int x;                                  \
346         arrtype* dst1 = dst;                    \
347                                                 \
348         for( x = 0; x < size.width; x += 3,     \
349                             dst1 += dststep )   \
350         {                                       \
351             arrtype t0 = src[x];                \
352             arrtype t1 = src[x + 1];            \
353             arrtype t2 = src[x + 2];            \
354                                                 \
355             dst1[0] = t0;                       \
356             dst1[1] = t1;                       \
357             dst1[2] = t2;                       \
358         }                                       \
359     }                                           \
360 }
361 
362 
363 #define ICV_DEF_TRANSP_CASE_C4( arrtype )       \
364 {                                               \
365     size.width *= 4;                            \
366     srcstep /= sizeof(src[0]);                  \
367     dststep /= sizeof(dst[0]);                  \
368                                                 \
369     for( ; size.height--; src+=srcstep, dst+=4 )\
370     {                                           \
371         int x;                                  \
372         arrtype* dst1 = dst;                    \
373                                                 \
374         for( x = 0; x < size.width; x += 4,     \
375                             dst1 += dststep )   \
376         {                                       \
377             arrtype t0 = src[x];                \
378             arrtype t1 = src[x + 1];            \
379                                                 \
380             dst1[0] = t0;                       \
381             dst1[1] = t1;                       \
382                                                 \
383             t0 = src[x + 2];                    \
384             t1 = src[x + 3];                    \
385                                                 \
386             dst1[2] = t0;                       \
387             dst1[3] = t1;                       \
388         }                                       \
389     }                                           \
390 }
391 
392 
393 #define ICV_DEF_TRANSP_INP_FUNC( flavor, arrtype, cn )      \
394 static CvStatus CV_STDCALL                                  \
395 icvTranspose_##flavor( arrtype* arr, int step, CvSize size )\
396 {                                                           \
397     assert( size.width == size.height );                    \
398                                                             \
399     ICV_DEF_TRANSP_INP_CASE_C##cn( arrtype, size.width )    \
400     return CV_OK;                                           \
401 }
402 
403 
404 #define ICV_DEF_TRANSP_FUNC( flavor, arrtype, cn )          \
405 static CvStatus CV_STDCALL                                  \
406 icvTranspose_##flavor( const arrtype* src, int srcstep,     \
407                     arrtype* dst, int dststep, CvSize size )\
408 {                                                           \
409     ICV_DEF_TRANSP_CASE_C##cn( arrtype )                    \
410     return CV_OK;                                           \
411 }
412 
413 
414 ICV_DEF_TRANSP_INP_FUNC( 8u_C1IR, uchar, 1 )
415 ICV_DEF_TRANSP_INP_FUNC( 8u_C2IR, ushort, 1 )
416 ICV_DEF_TRANSP_INP_FUNC( 8u_C3IR, uchar, 3 )
417 ICV_DEF_TRANSP_INP_FUNC( 16u_C2IR, int, 1 )
418 ICV_DEF_TRANSP_INP_FUNC( 16u_C3IR, ushort, 3 )
419 ICV_DEF_TRANSP_INP_FUNC( 32s_C2IR, int64, 1 )
420 ICV_DEF_TRANSP_INP_FUNC( 32s_C3IR, int, 3 )
421 ICV_DEF_TRANSP_INP_FUNC( 64s_C2IR, int, 4 )
422 ICV_DEF_TRANSP_INP_FUNC( 64s_C3IR, int64, 3 )
423 ICV_DEF_TRANSP_INP_FUNC( 64s_C4IR, int64, 4 )
424 
425 
426 ICV_DEF_TRANSP_FUNC( 8u_C1R, uchar, 1 )
427 ICV_DEF_TRANSP_FUNC( 8u_C2R, ushort, 1 )
428 ICV_DEF_TRANSP_FUNC( 8u_C3R, uchar, 3 )
429 ICV_DEF_TRANSP_FUNC( 16u_C2R, int, 1 )
430 ICV_DEF_TRANSP_FUNC( 16u_C3R, ushort, 3 )
431 ICV_DEF_TRANSP_FUNC( 32s_C2R, int64, 1 )
432 ICV_DEF_TRANSP_FUNC( 32s_C3R, int, 3 )
433 ICV_DEF_TRANSP_FUNC( 64s_C2R, int, 4 )
434 ICV_DEF_TRANSP_FUNC( 64s_C3R, int64, 3 )
435 ICV_DEF_TRANSP_FUNC( 64s_C4R, int64, 4 )
436 
CV_DEF_INIT_PIXSIZE_TAB_2D(Transpose,R)437 CV_DEF_INIT_PIXSIZE_TAB_2D( Transpose, R )
438 CV_DEF_INIT_PIXSIZE_TAB_2D( Transpose, IR )
439 
440 CV_IMPL void
441 cvTranspose( const CvArr* srcarr, CvArr* dstarr )
442 {
443     static CvBtFuncTable tab, inp_tab;
444     static int inittab = 0;
445 
446     CV_FUNCNAME( "cvTranspose" );
447 
448     __BEGIN__;
449 
450     CvMat sstub, *src = (CvMat*)srcarr;
451     CvMat dstub, *dst = (CvMat*)dstarr;
452     CvSize size;
453     int type, pix_size;
454 
455     if( !inittab )
456     {
457         icvInitTransposeIRTable( &inp_tab );
458         icvInitTransposeRTable( &tab );
459         inittab = 1;
460     }
461 
462     if( !CV_IS_MAT( src ))
463     {
464         int coi = 0;
465         CV_CALL( src = cvGetMat( src, &sstub, &coi ));
466         if( coi != 0 )
467             CV_ERROR( CV_BadCOI, "coi is not supported" );
468     }
469 
470     type = CV_MAT_TYPE( src->type );
471     pix_size = CV_ELEM_SIZE(type);
472     size = cvGetMatSize( src );
473 
474     if( dstarr == srcarr )
475     {
476         dst = src;
477     }
478     else
479     {
480         if( !CV_IS_MAT( dst ))
481         {
482             int coi = 0;
483             CV_CALL( dst = cvGetMat( dst, &dstub, &coi ));
484 
485             if( coi != 0 )
486             CV_ERROR( CV_BadCOI, "coi is not supported" );
487         }
488 
489         if( !CV_ARE_TYPES_EQ( src, dst ))
490             CV_ERROR( CV_StsUnmatchedFormats, "" );
491 
492         if( size.width != dst->height || size.height != dst->width )
493             CV_ERROR( CV_StsUnmatchedSizes, "" );
494     }
495 
496     if( src->data.ptr == dst->data.ptr )
497     {
498         if( size.width == size.height )
499         {
500             CvFunc2D_1A func = (CvFunc2D_1A)(inp_tab.fn_2d[pix_size]);
501 
502             if( !func )
503                 CV_ERROR( CV_StsUnsupportedFormat, "" );
504 
505             IPPI_CALL( func( src->data.ptr, src->step, size ));
506         }
507         else
508         {
509             if( size.width != 1 && size.height != 1 )
510                 CV_ERROR( CV_StsBadSize,
511                     "Rectangular matrix can not be transposed inplace" );
512 
513             if( !CV_IS_MAT_CONT( src->type & dst->type ))
514                 CV_ERROR( CV_StsBadFlag, "In case of inplace column/row transposition "
515                                        "both source and destination must be continuous" );
516 
517             if( dst == src )
518             {
519                 int t;
520                 CV_SWAP( dst->width, dst->height, t );
521                 dst->step = dst->height == 1 ? 0 : pix_size;
522             }
523         }
524     }
525     else
526     {
527         CvFunc2D_2A func = (CvFunc2D_2A)(tab.fn_2d[pix_size]);
528 
529         if( !func )
530             CV_ERROR( CV_StsUnsupportedFormat, "" );
531 
532         IPPI_CALL( func( src->data.ptr, src->step,
533                          dst->data.ptr, dst->step, size ));
534     }
535 
536     __END__;
537 }
538 
539 
540 /****************************************************************************************\
541 *                              LU decomposition/back substitution                        *
542 \****************************************************************************************/
543 
544 CV_IMPL void
cvCompleteSymm(CvMat * matrix,int LtoR)545 cvCompleteSymm( CvMat* matrix, int LtoR )
546 {
547     CV_FUNCNAME( "cvCompleteSymm" );
548 
549     __BEGIN__;
550 
551     int i, j, nrows;
552 
553     CV_ASSERT( CV_IS_MAT(matrix) && matrix->rows == matrix->cols );
554 
555     nrows = matrix->rows;
556 
557     if( CV_MAT_TYPE(matrix->type) == CV_32FC1 || CV_MAT_TYPE(matrix->type) == CV_32SC1 )
558     {
559         int* data = matrix->data.i;
560         int step = matrix->step/sizeof(data[0]);
561         int j0 = 0, j1 = nrows;
562         for( i = 0; i < nrows; i++ )
563         {
564             if( !LtoR ) j1 = i; else j0 = i+1;
565             for( j = j0; j < j1; j++ )
566                 data[i*step + j] = data[j*step + i];
567         }
568     }
569     else if( CV_MAT_TYPE(matrix->type) == CV_64FC1 )
570     {
571         double* data = matrix->data.db;
572         int step = matrix->step/sizeof(data[0]);
573         int j0 = 0, j1 = nrows;
574         for( i = 0; i < nrows; i++ )
575         {
576             if( !LtoR ) j1 = i; else j0 = i+1;
577             for( j = j0; j < j1; j++ )
578                 data[i*step + j] = data[j*step + i];
579         }
580     }
581     else
582         CV_ERROR( CV_StsUnsupportedFormat, "" );
583 
584     __END__;
585 }
586 
587 
588 /****************************************************************************************\
589 *                              LU decomposition/back substitution                        *
590 \****************************************************************************************/
591 
592 #define arrtype float
593 #define temptype double
594 
595 typedef  CvStatus (CV_STDCALL * CvLUDecompFunc)( double* A, int stepA, CvSize sizeA,
596                                                  void* B, int stepB, CvSize sizeB,
597                                                  double* det );
598 
599 typedef  CvStatus (CV_STDCALL * CvLUBackFunc)( double* A, int stepA, CvSize sizeA,
600                                                void* B, int stepB, CvSize sizeB );
601 
602 
603 #define ICV_DEF_LU_DECOMP_FUNC( flavor, arrtype )                               \
604 static CvStatus CV_STDCALL                                                      \
605 icvLUDecomp_##flavor( double* A, int stepA, CvSize sizeA,                       \
606                       arrtype* B, int stepB, CvSize sizeB, double* _det )       \
607 {                                                                               \
608     int n = sizeA.width;                                                        \
609     int m = 0, i;                                                               \
610     double det = 1;                                                             \
611                                                                                 \
612     assert( sizeA.width == sizeA.height );                                      \
613                                                                                 \
614     if( B )                                                                     \
615     {                                                                           \
616         assert( sizeA.height == sizeB.height );                                 \
617         m = sizeB.width;                                                        \
618     }                                                                           \
619     stepA /= sizeof(A[0]);                                                      \
620     stepB /= sizeof(B[0]);                                                      \
621                                                                                 \
622     for( i = 0; i < n; i++, A += stepA, B += stepB )                            \
623     {                                                                           \
624         int j, k = i;                                                           \
625         double* tA = A;                                                         \
626         arrtype* tB = 0;                                                        \
627         double kval = fabs(A[i]), tval;                                         \
628                                                                                 \
629         /* find the pivot element */                                            \
630         for( j = i + 1; j < n; j++ )                                            \
631         {                                                                       \
632             tA += stepA;                                                        \
633             tval = fabs(tA[i]);                                                 \
634                                                                                 \
635             if( tval > kval )                                                   \
636             {                                                                   \
637                 kval = tval;                                                    \
638                 k = j;                                                          \
639             }                                                                   \
640         }                                                                       \
641                                                                                 \
642         if( kval == 0 )                                                         \
643         {                                                                       \
644             det = 0;                                                            \
645             break;                                                              \
646         }                                                                       \
647                                                                                 \
648         /* swap rows */                                                         \
649         if( k != i )                                                            \
650         {                                                                       \
651             tA = A + stepA*(k - i);                                             \
652             det = -det;                                                         \
653                                                                                 \
654             for( j = i; j < n; j++ )                                            \
655             {                                                                   \
656                 double t;                                                       \
657                 CV_SWAP( A[j], tA[j], t );                                      \
658             }                                                                   \
659                                                                                 \
660             if( m > 0 )                                                         \
661             {                                                                   \
662                 tB = B + stepB*(k - i);                                         \
663                                                                                 \
664                 for( j = 0; j < m; j++ )                                        \
665                 {                                                               \
666                     arrtype t = B[j];                                           \
667                     CV_SWAP( B[j], tB[j], t );                                  \
668                 }                                                               \
669             }                                                                   \
670         }                                                                       \
671                                                                                 \
672         tval = 1./A[i];                                                         \
673         det *= A[i];                                                            \
674         tA = A;                                                                 \
675         tB = B;                                                                 \
676         A[i] = tval; /* to replace division with multiplication in LUBack */    \
677                                                                                 \
678         /* update matrix and the right side of the system */                    \
679         for( j = i + 1; j < n; j++ )                                            \
680         {                                                                       \
681             tA += stepA;                                                        \
682             tB += stepB;                                                        \
683             double alpha = -tA[i]*tval;                                         \
684                                                                                 \
685             for( k = i + 1; k < n; k++ )                                        \
686                 tA[k] = tA[k] + alpha*A[k];                                     \
687                                                                                 \
688             if( m > 0 )                                                         \
689                 for( k = 0; k < m; k++ )                                        \
690                     tB[k] = (arrtype)(tB[k] + alpha*B[k]);                      \
691         }                                                                       \
692     }                                                                           \
693                                                                                 \
694     if( _det )                                                                  \
695         *_det = det;                                                            \
696                                                                                 \
697     return CV_OK;                                                               \
698 }
699 
700 
701 ICV_DEF_LU_DECOMP_FUNC( 32f, float )
702 ICV_DEF_LU_DECOMP_FUNC( 64f, double )
703 
704 
705 #define ICV_DEF_LU_BACK_FUNC( flavor, arrtype )                                 \
706 static CvStatus CV_STDCALL                                                      \
707 icvLUBack_##flavor( double* A, int stepA, CvSize sizeA,                         \
708                     arrtype* B, int stepB, CvSize sizeB )                       \
709 {                                                                               \
710     int n = sizeA.width;                                                        \
711     int m = sizeB.width, i;                                                     \
712                                                                                 \
713     assert( m > 0 && sizeA.width == sizeA.height &&                             \
714             sizeA.height == sizeB.height );                                     \
715     stepA /= sizeof(A[0]);                                                      \
716     stepB /= sizeof(B[0]);                                                      \
717                                                                                 \
718     A += stepA*(n - 1);                                                         \
719     B += stepB*(n - 1);                                                         \
720                                                                                 \
721     for( i = n - 1; i >= 0; i--, A -= stepA )                                   \
722     {                                                                           \
723         int j, k;                                                               \
724         for( j = 0; j < m; j++ )                                                \
725         {                                                                       \
726             arrtype* tB = B + j;                                                \
727             double x = 0;                                                       \
728                                                                                 \
729             for( k = n - 1; k > i; k--, tB -= stepB )                           \
730                 x += A[k]*tB[0];                                                \
731                                                                                 \
732             tB[0] = (arrtype)((tB[0] - x)*A[i]);                                \
733         }                                                                       \
734     }                                                                           \
735                                                                                 \
736     return CV_OK;                                                               \
737 }
738 
739 
740 ICV_DEF_LU_BACK_FUNC( 32f, float )
741 ICV_DEF_LU_BACK_FUNC( 64f, double )
742 
743 static CvFuncTable lu_decomp_tab, lu_back_tab;
744 static int lu_inittab = 0;
745 
icvInitLUTable(CvFuncTable * decomp_tab,CvFuncTable * back_tab)746 static void icvInitLUTable( CvFuncTable* decomp_tab,
747                             CvFuncTable* back_tab )
748 {
749     decomp_tab->fn_2d[0] = (void*)icvLUDecomp_32f;
750     decomp_tab->fn_2d[1] = (void*)icvLUDecomp_64f;
751     back_tab->fn_2d[0] = (void*)icvLUBack_32f;
752     back_tab->fn_2d[1] = (void*)icvLUBack_64f;
753 }
754 
755 
756 
757 /****************************************************************************************\
758 *                                 Determinant of the matrix                              *
759 \****************************************************************************************/
760 
761 #define det2(m)   (m(0,0)*m(1,1) - m(0,1)*m(1,0))
762 #define det3(m)   (m(0,0)*(m(1,1)*m(2,2) - m(1,2)*m(2,1)) -  \
763                    m(0,1)*(m(1,0)*m(2,2) - m(1,2)*m(2,0)) +  \
764                    m(0,2)*(m(1,0)*m(2,1) - m(1,1)*m(2,0)))
765 
766 CV_IMPL double
cvDet(const CvArr * arr)767 cvDet( const CvArr* arr )
768 {
769     double result = 0;
770     uchar* buffer = 0;
771     int local_alloc = 0;
772 
773     CV_FUNCNAME( "cvDet" );
774 
775     __BEGIN__;
776 
777     CvMat stub, *mat = (CvMat*)arr;
778     int type;
779 
780     if( !CV_IS_MAT( mat ))
781     {
782         CV_CALL( mat = cvGetMat( mat, &stub ));
783     }
784 
785     type = CV_MAT_TYPE( mat->type );
786 
787     if( mat->width != mat->height )
788         CV_ERROR( CV_StsBadSize, "The matrix must be square" );
789 
790     #define Mf( y, x ) ((float*)(m + y*step))[x]
791     #define Md( y, x ) ((double*)(m + y*step))[x]
792 
793     if( mat->width == 2 )
794     {
795         uchar* m = mat->data.ptr;
796         int step = mat->step;
797 
798         if( type == CV_32FC1 )
799         {
800             result = det2(Mf);
801         }
802         else if( type == CV_64FC1 )
803         {
804             result = det2(Md);
805         }
806         else
807         {
808             CV_ERROR( CV_StsUnsupportedFormat, "" );
809         }
810     }
811     else if( mat->width == 3 )
812     {
813         uchar* m = mat->data.ptr;
814         int step = mat->step;
815 
816         if( type == CV_32FC1 )
817         {
818             result = det3(Mf);
819         }
820         else if( type == CV_64FC1 )
821         {
822             result = det3(Md);
823         }
824         else
825         {
826             CV_ERROR( CV_StsUnsupportedFormat, "" );
827         }
828     }
829     else if( mat->width == 1 )
830     {
831         if( type == CV_32FC1 )
832         {
833             result = mat->data.fl[0];
834         }
835         else if( type == CV_64FC1 )
836         {
837             result = mat->data.db[0];
838         }
839         else
840         {
841             CV_ERROR( CV_StsUnsupportedFormat, "" );
842         }
843     }
844     else
845     {
846         CvLUDecompFunc decomp_func;
847         CvSize size = cvGetMatSize( mat );
848         const int worktype = CV_64FC1;
849         int buf_size = size.width*size.height*CV_ELEM_SIZE(worktype);
850         CvMat tmat;
851 
852         if( !lu_inittab )
853         {
854             icvInitLUTable( &lu_decomp_tab, &lu_back_tab );
855             lu_inittab = 1;
856         }
857 
858         if( CV_MAT_CN( type ) != 1 || CV_MAT_DEPTH( type ) < CV_32F )
859             CV_ERROR( CV_StsUnsupportedFormat, "" );
860 
861         if( size.width <= CV_MAX_LOCAL_MAT_SIZE )
862         {
863             buffer = (uchar*)cvStackAlloc( buf_size );
864             local_alloc = 1;
865         }
866         else
867         {
868             CV_CALL( buffer = (uchar*)cvAlloc( buf_size ));
869         }
870 
871         CV_CALL( cvInitMatHeader( &tmat, size.height, size.width, worktype, buffer ));
872         if( type == worktype )
873         {
874         	CV_CALL( cvCopy( mat, &tmat ));
875         }
876         else
877             CV_CALL( cvConvert( mat, &tmat ));
878 
879         decomp_func = (CvLUDecompFunc)(lu_decomp_tab.fn_2d[CV_MAT_DEPTH(worktype)-CV_32F]);
880         assert( decomp_func );
881 
882         IPPI_CALL( decomp_func( tmat.data.db, tmat.step, size, 0, 0, size, &result ));
883     }
884 
885     #undef Mf
886     #undef Md
887 
888     /*icvCheckVector_64f( &result, 1 );*/
889 
890     __END__;
891 
892     if( buffer && !local_alloc )
893         cvFree( &buffer );
894 
895     return result;
896 }
897 
898 
899 
900 /****************************************************************************************\
901 *                          Inverse (or pseudo-inverse) of the matrix                     *
902 \****************************************************************************************/
903 
904 #define Sf( y, x ) ((float*)(srcdata + y*srcstep))[x]
905 #define Sd( y, x ) ((double*)(srcdata + y*srcstep))[x]
906 #define Df( y, x ) ((float*)(dstdata + y*dststep))[x]
907 #define Dd( y, x ) ((double*)(dstdata + y*dststep))[x]
908 
909 CV_IMPL double
cvInvert(const CvArr * srcarr,CvArr * dstarr,int method)910 cvInvert( const CvArr* srcarr, CvArr* dstarr, int method )
911 {
912     CvMat* u = 0;
913     CvMat* v = 0;
914     CvMat* w = 0;
915 
916     uchar* buffer = 0;
917     int local_alloc = 0;
918     double result = 0;
919 
920     CV_FUNCNAME( "cvInvert" );
921 
922     __BEGIN__;
923 
924     CvMat sstub, *src = (CvMat*)srcarr;
925     CvMat dstub, *dst = (CvMat*)dstarr;
926     int type;
927 
928     if( !CV_IS_MAT( src ))
929         CV_CALL( src = cvGetMat( src, &sstub ));
930 
931     if( !CV_IS_MAT( dst ))
932         CV_CALL( dst = cvGetMat( dst, &dstub ));
933 
934     type = CV_MAT_TYPE( src->type );
935 
936     if( method == CV_SVD || method == CV_SVD_SYM )
937     {
938         int n = MIN(src->rows,src->cols);
939         if( method == CV_SVD_SYM && src->rows != src->cols )
940             CV_ERROR( CV_StsBadSize, "CV_SVD_SYM method is used for non-square matrix" );
941 
942         CV_CALL( u = cvCreateMat( n, src->rows, src->type ));
943         if( method != CV_SVD_SYM )
944             CV_CALL( v = cvCreateMat( n, src->cols, src->type ));
945         CV_CALL( w = cvCreateMat( n, 1, src->type ));
946         CV_CALL( cvSVD( src, w, u, v, CV_SVD_U_T + CV_SVD_V_T ));
947 
948         if( type == CV_32FC1 )
949             result = w->data.fl[0] >= FLT_EPSILON ?
950                      w->data.fl[w->rows-1]/w->data.fl[0] : 0;
951         else
952             result = w->data.db[0] >= FLT_EPSILON ?
953                      w->data.db[w->rows-1]/w->data.db[0] : 0;
954 
955         CV_CALL( cvSVBkSb( w, u, v ? v : u, 0, dst, CV_SVD_U_T + CV_SVD_V_T ));
956         EXIT;
957     }
958     else if( method != CV_LU )
959         CV_ERROR( CV_StsBadArg, "Unknown inversion method" );
960 
961     if( !CV_ARE_TYPES_EQ( src, dst ))
962         CV_ERROR( CV_StsUnmatchedFormats, "" );
963 
964     if( src->width != src->height )
965         CV_ERROR( CV_StsBadSize, "The matrix must be square" );
966 
967     if( !CV_ARE_SIZES_EQ( src, dst ))
968         CV_ERROR( CV_StsUnmatchedSizes, "" );
969 
970     if( type != CV_32FC1 && type != CV_64FC1 )
971         CV_ERROR( CV_StsUnsupportedFormat, "" );
972 
973     if( src->width <= 3 )
974     {
975         uchar* srcdata = src->data.ptr;
976         uchar* dstdata = dst->data.ptr;
977         int srcstep = src->step;
978         int dststep = dst->step;
979 
980         if( src->width == 2 )
981         {
982             if( type == CV_32FC1 )
983             {
984                 double d = det2(Sf);
985                 if( d != 0. )
986                 {
987                     double t0, t1;
988                     result = d;
989                     d = 1./d;
990                     t0 = Sf(0,0)*d;
991                     t1 = Sf(1,1)*d;
992                     Df(1,1) = (float)t0;
993                     Df(0,0) = (float)t1;
994                     t0 = -Sf(0,1)*d;
995                     t1 = -Sf(1,0)*d;
996                     Df(0,1) = (float)t0;
997                     Df(1,0) = (float)t1;
998                 }
999             }
1000             else
1001             {
1002                 double d = det2(Sd);
1003                 if( d != 0. )
1004                 {
1005                     double t0, t1;
1006                     result = d;
1007                     d = 1./d;
1008                     t0 = Sd(0,0)*d;
1009                     t1 = Sd(1,1)*d;
1010                     Dd(1,1) = t0;
1011                     Dd(0,0) = t1;
1012                     t0 = -Sd(0,1)*d;
1013                     t1 = -Sd(1,0)*d;
1014                     Dd(0,1) = t0;
1015                     Dd(1,0) = t1;
1016                 }
1017             }
1018         }
1019         else if( src->width == 3 )
1020         {
1021             if( type == CV_32FC1 )
1022             {
1023                 double d = det3(Sf);
1024                 if( d != 0. )
1025                 {
1026                     float t[9];
1027                     result = d;
1028                     d = 1./d;
1029 
1030                     t[0] = (float)((Sf(1,1) * Sf(2,2) - Sf(1,2) * Sf(2,1)) * d);
1031                     t[1] = (float)((Sf(0,2) * Sf(2,1) - Sf(0,1) * Sf(2,2)) * d);
1032                     t[2] = (float)((Sf(0,1) * Sf(1,2) - Sf(0,2) * Sf(1,1)) * d);
1033 
1034                     t[3] = (float)((Sf(1,2) * Sf(2,0) - Sf(1,0) * Sf(2,2)) * d);
1035                     t[4] = (float)((Sf(0,0) * Sf(2,2) - Sf(0,2) * Sf(2,0)) * d);
1036                     t[5] = (float)((Sf(0,2) * Sf(1,0) - Sf(0,0) * Sf(1,2)) * d);
1037 
1038                     t[6] = (float)((Sf(1,0) * Sf(2,1) - Sf(1,1) * Sf(2,0)) * d);
1039                     t[7] = (float)((Sf(0,1) * Sf(2,0) - Sf(0,0) * Sf(2,1)) * d);
1040                     t[8] = (float)((Sf(0,0) * Sf(1,1) - Sf(0,1) * Sf(1,0)) * d);
1041 
1042                     Df(0,0) = t[0]; Df(0,1) = t[1]; Df(0,2) = t[2];
1043                     Df(1,0) = t[3]; Df(1,1) = t[4]; Df(1,2) = t[5];
1044                     Df(2,0) = t[6]; Df(2,1) = t[7]; Df(2,2) = t[8];
1045                 }
1046             }
1047             else
1048             {
1049                 double d = det3(Sd);
1050                 if( d != 0. )
1051                 {
1052                     double t[9];
1053                     result = d;
1054                     d = 1./d;
1055 
1056                     t[0] = (Sd(1,1) * Sd(2,2) - Sd(1,2) * Sd(2,1)) * d;
1057                     t[1] = (Sd(0,2) * Sd(2,1) - Sd(0,1) * Sd(2,2)) * d;
1058                     t[2] = (Sd(0,1) * Sd(1,2) - Sd(0,2) * Sd(1,1)) * d;
1059 
1060                     t[3] = (Sd(1,2) * Sd(2,0) - Sd(1,0) * Sd(2,2)) * d;
1061                     t[4] = (Sd(0,0) * Sd(2,2) - Sd(0,2) * Sd(2,0)) * d;
1062                     t[5] = (Sd(0,2) * Sd(1,0) - Sd(0,0) * Sd(1,2)) * d;
1063 
1064                     t[6] = (Sd(1,0) * Sd(2,1) - Sd(1,1) * Sd(2,0)) * d;
1065                     t[7] = (Sd(0,1) * Sd(2,0) - Sd(0,0) * Sd(2,1)) * d;
1066                     t[8] = (Sd(0,0) * Sd(1,1) - Sd(0,1) * Sd(1,0)) * d;
1067 
1068                     Dd(0,0) = t[0]; Dd(0,1) = t[1]; Dd(0,2) = t[2];
1069                     Dd(1,0) = t[3]; Dd(1,1) = t[4]; Dd(1,2) = t[5];
1070                     Dd(2,0) = t[6]; Dd(2,1) = t[7]; Dd(2,2) = t[8];
1071                 }
1072             }
1073         }
1074         else
1075         {
1076             assert( src->width == 1 );
1077 
1078             if( type == CV_32FC1 )
1079             {
1080                 double d = Sf(0,0);
1081                 if( d != 0. )
1082                 {
1083                     result = d;
1084                     Df(0,0) = (float)(1./d);
1085                 }
1086             }
1087             else
1088             {
1089                 double d = Sd(0,0);
1090                 if( d != 0. )
1091                 {
1092                     result = d;
1093                     Dd(0,0) = 1./d;
1094                 }
1095             }
1096         }
1097     }
1098     else
1099     {
1100         CvLUDecompFunc decomp_func;
1101         CvLUBackFunc back_func;
1102         CvSize size = cvGetMatSize( src );
1103         const int worktype = CV_64FC1;
1104         int buf_size = size.width*size.height*CV_ELEM_SIZE(worktype);
1105         CvMat tmat;
1106 
1107         if( !lu_inittab )
1108         {
1109             icvInitLUTable( &lu_decomp_tab, &lu_back_tab );
1110             lu_inittab = 1;
1111         }
1112 
1113         if( size.width <= CV_MAX_LOCAL_MAT_SIZE )
1114         {
1115             buffer = (uchar*)cvStackAlloc( buf_size );
1116             local_alloc = 1;
1117         }
1118         else
1119         {
1120             CV_CALL( buffer = (uchar*)cvAlloc( buf_size ));
1121         }
1122 
1123         CV_CALL( cvInitMatHeader( &tmat, size.height, size.width, worktype, buffer ));
1124         if( type == worktype )
1125         {
1126             CV_CALL( cvCopy( src, &tmat ));
1127         }
1128         else
1129             CV_CALL( cvConvert( src, &tmat ));
1130         CV_CALL( cvSetIdentity( dst ));
1131 
1132         decomp_func = (CvLUDecompFunc)(lu_decomp_tab.fn_2d[CV_MAT_DEPTH(type)-CV_32F]);
1133         back_func = (CvLUBackFunc)(lu_back_tab.fn_2d[CV_MAT_DEPTH(type)-CV_32F]);
1134         assert( decomp_func && back_func );
1135 
1136         IPPI_CALL( decomp_func( tmat.data.db, tmat.step, size,
1137                                 dst->data.ptr, dst->step, size, &result ));
1138 
1139         if( result != 0 )
1140         {
1141             IPPI_CALL( back_func( tmat.data.db, tmat.step, size,
1142                                   dst->data.ptr, dst->step, size ));
1143         }
1144     }
1145 
1146     if( !result )
1147         CV_CALL( cvSetZero( dst ));
1148 
1149     __END__;
1150 
1151     if( buffer && !local_alloc )
1152         cvFree( &buffer );
1153 
1154     if( u || v || w )
1155     {
1156         cvReleaseMat( &u );
1157         cvReleaseMat( &v );
1158         cvReleaseMat( &w );
1159     }
1160 
1161     return result;
1162 }
1163 
1164 
1165 /****************************************************************************************\
1166 *                               Linear system [least-squares] solution                   *
1167 \****************************************************************************************/
1168 
1169 static void
icvLSQ(const CvMat * A,const CvMat * B,CvMat * X)1170 icvLSQ( const CvMat* A, const CvMat* B, CvMat* X )
1171 {
1172     CvMat* AtA = 0;
1173     CvMat* AtB = 0;
1174     CvMat* W = 0;
1175     CvMat* V = 0;
1176 
1177     CV_FUNCNAME( "icvLSQ" );
1178 
1179     __BEGIN__;
1180 
1181     if( !CV_IS_MAT(A) || !CV_IS_MAT(B) || !CV_IS_MAT(X) )
1182         CV_ERROR( CV_StsBadArg, "Some of required arguments is not a valid matrix" );
1183 
1184     AtA = cvCreateMat( A->cols, A->cols, A->type );
1185     AtB = cvCreateMat( A->cols, 1, A->type );
1186     W = cvCreateMat( A->cols, 1, A->type );
1187     V = cvCreateMat( A->cols, A->cols, A->type );
1188 
1189     cvMulTransposed( A, AtA, 1 );
1190     cvGEMM( A, B, 1, 0, 0, AtB, CV_GEMM_A_T );
1191     cvSVD( AtA, W, 0, V, CV_SVD_MODIFY_A + CV_SVD_V_T );
1192     cvSVBkSb( W, V, V, AtB, X, CV_SVD_U_T + CV_SVD_V_T );
1193 
1194     __END__;
1195 
1196     cvReleaseMat( &AtA );
1197     cvReleaseMat( &AtB );
1198     cvReleaseMat( &W );
1199     cvReleaseMat( &V );
1200 }
1201 
1202 CV_IMPL int
cvSolve(const CvArr * A,const CvArr * b,CvArr * x,int method)1203 cvSolve( const CvArr* A, const CvArr* b, CvArr* x, int method )
1204 {
1205     CvMat* u = 0;
1206     CvMat* v = 0;
1207     CvMat* w = 0;
1208 
1209     uchar* buffer = 0;
1210     int local_alloc = 0;
1211     int result = 1;
1212 
1213     CV_FUNCNAME( "cvSolve" );
1214 
1215     __BEGIN__;
1216 
1217     CvMat sstub, *src = (CvMat*)A;
1218     CvMat dstub, *dst = (CvMat*)x;
1219     CvMat bstub, *src2 = (CvMat*)b;
1220     int type;
1221 
1222     if( !CV_IS_MAT( src ))
1223         CV_CALL( src = cvGetMat( src, &sstub ));
1224 
1225     if( !CV_IS_MAT( src2 ))
1226         CV_CALL( src2 = cvGetMat( src2, &bstub ));
1227 
1228     if( !CV_IS_MAT( dst ))
1229         CV_CALL( dst = cvGetMat( dst, &dstub ));
1230 
1231     if( method & CV_LSQ )
1232     {
1233         icvLSQ( src, src2, dst );
1234         EXIT;
1235     }
1236 
1237     if( method == CV_SVD || method == CV_SVD_SYM )
1238     {
1239         int n = MIN(src->rows,src->cols);
1240 
1241         if( method == CV_SVD_SYM && src->rows != src->cols )
1242             CV_ERROR( CV_StsBadSize, "CV_SVD_SYM method is used for non-square matrix" );
1243 
1244         CV_CALL( u = cvCreateMat( n, src->rows, src->type ));
1245         if( method != CV_SVD_SYM )
1246             CV_CALL( v = cvCreateMat( n, src->cols, src->type ));
1247         CV_CALL( w = cvCreateMat( n, 1, src->type ));
1248         CV_CALL( cvSVD( src, w, u, v, CV_SVD_U_T + CV_SVD_V_T ));
1249         CV_CALL( cvSVBkSb( w, u, v ? v : u, src2, dst, CV_SVD_U_T + CV_SVD_V_T ));
1250         EXIT;
1251     }
1252     else if( method != CV_LU )
1253         CV_ERROR( CV_StsBadArg, "Unknown inversion method" );
1254 
1255     type = CV_MAT_TYPE( src->type );
1256 
1257     if( !CV_ARE_TYPES_EQ( src, dst ) || !CV_ARE_TYPES_EQ( src, src2 ))
1258         CV_ERROR( CV_StsUnmatchedFormats, "" );
1259 
1260     if( src->width != src->height )
1261         CV_ERROR( CV_StsBadSize, "The matrix must be square" );
1262 
1263     if( !CV_ARE_SIZES_EQ( src2, dst ) || src->width != src2->height )
1264         CV_ERROR( CV_StsUnmatchedSizes, "" );
1265 
1266     if( type != CV_32FC1 && type != CV_64FC1 )
1267         CV_ERROR( CV_StsUnsupportedFormat, "" );
1268 
1269     // check case of a single equation and small matrix
1270     if( src->width <= 3 && src2->width == 1 )
1271     {
1272         #define bf(y) ((float*)(bdata + y*src2step))[0]
1273         #define bd(y) ((double*)(bdata + y*src2step))[0]
1274 
1275         uchar* srcdata = src->data.ptr;
1276         uchar* bdata = src2->data.ptr;
1277         uchar* dstdata = dst->data.ptr;
1278         int srcstep = src->step;
1279         int src2step = src2->step;
1280         int dststep = dst->step;
1281 
1282         if( src->width == 2 )
1283         {
1284             if( type == CV_32FC1 )
1285             {
1286                 double d = det2(Sf);
1287                 if( d != 0. )
1288                 {
1289                     float t;
1290                     d = 1./d;
1291                     t = (float)((bf(0)*Sf(1,1) - bf(1)*Sf(0,1))*d);
1292                     Df(1,0) = (float)((bf(1)*Sf(0,0) - bf(0)*Sf(1,0))*d);
1293                     Df(0,0) = t;
1294                 }
1295                 else
1296                     result = 0;
1297             }
1298             else
1299             {
1300                 double d = det2(Sd);
1301                 if( d != 0. )
1302                 {
1303                     double t;
1304                     d = 1./d;
1305                     t = (bd(0)*Sd(1,1) - bd(1)*Sd(0,1))*d;
1306                     Dd(1,0) = (bd(1)*Sd(0,0) - bd(0)*Sd(1,0))*d;
1307                     Dd(0,0) = t;
1308                 }
1309                 else
1310                     result = 0;
1311             }
1312         }
1313         else if( src->width == 3 )
1314         {
1315             if( type == CV_32FC1 )
1316             {
1317                 double d = det3(Sf);
1318                 if( d != 0. )
1319                 {
1320                     float t[3];
1321                     d = 1./d;
1322 
1323                     t[0] = (float)(d*
1324                            (bf(0)*(Sf(1,1)*Sf(2,2) - Sf(1,2)*Sf(2,1)) -
1325                             Sf(0,1)*(bf(1)*Sf(2,2) - Sf(1,2)*bf(2)) +
1326                             Sf(0,2)*(bf(1)*Sf(2,1) - Sf(1,1)*bf(2))));
1327 
1328                     t[1] = (float)(d*
1329                            (Sf(0,0)*(bf(1)*Sf(2,2) - Sf(1,2)*bf(2)) -
1330                             bf(0)*(Sf(1,0)*Sf(2,2) - Sf(1,2)*Sf(2,0)) +
1331                             Sf(0,2)*(Sf(1,0)*bf(2) - bf(1)*Sf(2,0))));
1332 
1333                     t[2] = (float)(d*
1334                            (Sf(0,0)*(Sf(1,1)*bf(2) - bf(1)*Sf(2,1)) -
1335                             Sf(0,1)*(Sf(1,0)*bf(2) - bf(1)*Sf(2,0)) +
1336                             bf(0)*(Sf(1,0)*Sf(2,1) - Sf(1,1)*Sf(2,0))));
1337 
1338                     Df(0,0) = t[0];
1339                     Df(1,0) = t[1];
1340                     Df(2,0) = t[2];
1341                 }
1342                 else
1343                     result = 0;
1344             }
1345             else
1346             {
1347                 double d = det3(Sd);
1348                 if( d != 0. )
1349                 {
1350                     double t[9];
1351 
1352                     d = 1./d;
1353 
1354                     t[0] = ((Sd(1,1) * Sd(2,2) - Sd(1,2) * Sd(2,1))*bd(0) +
1355                             (Sd(0,2) * Sd(2,1) - Sd(0,1) * Sd(2,2))*bd(1) +
1356                             (Sd(0,1) * Sd(1,2) - Sd(0,2) * Sd(1,1))*bd(2))*d;
1357 
1358                     t[1] = ((Sd(1,2) * Sd(2,0) - Sd(1,0) * Sd(2,2))*bd(0) +
1359                             (Sd(0,0) * Sd(2,2) - Sd(0,2) * Sd(2,0))*bd(1) +
1360                             (Sd(0,2) * Sd(1,0) - Sd(0,0) * Sd(1,2))*bd(2))*d;
1361 
1362                     t[2] = ((Sd(1,0) * Sd(2,1) - Sd(1,1) * Sd(2,0))*bd(0) +
1363                             (Sd(0,1) * Sd(2,0) - Sd(0,0) * Sd(2,1))*bd(1) +
1364                             (Sd(0,0) * Sd(1,1) - Sd(0,1) * Sd(1,0))*bd(2))*d;
1365 
1366                     Dd(0,0) = t[0];
1367                     Dd(1,0) = t[1];
1368                     Dd(2,0) = t[2];
1369                 }
1370                 else
1371                     result = 0;
1372             }
1373         }
1374         else
1375         {
1376             assert( src->width == 1 );
1377 
1378             if( type == CV_32FC1 )
1379             {
1380                 double d = Sf(0,0);
1381                 if( d != 0. )
1382                     Df(0,0) = (float)(bf(0)/d);
1383                 else
1384                     result = 0;
1385             }
1386             else
1387             {
1388                 double d = Sd(0,0);
1389                 if( d != 0. )
1390                     Dd(0,0) = (bd(0)/d);
1391                 else
1392                     result = 0;
1393             }
1394         }
1395     }
1396     else
1397     {
1398         CvLUDecompFunc decomp_func;
1399         CvLUBackFunc back_func;
1400         CvSize size = cvGetMatSize( src );
1401         CvSize dstsize = cvGetMatSize( dst );
1402         int worktype = CV_64FC1;
1403         int buf_size = size.width*size.height*CV_ELEM_SIZE(worktype);
1404         double d = 0;
1405         CvMat tmat;
1406 
1407         if( !lu_inittab )
1408         {
1409             icvInitLUTable( &lu_decomp_tab, &lu_back_tab );
1410             lu_inittab = 1;
1411         }
1412 
1413         if( size.width <= CV_MAX_LOCAL_MAT_SIZE )
1414         {
1415             buffer = (uchar*)cvStackAlloc( buf_size );
1416             local_alloc = 1;
1417         }
1418         else
1419         {
1420             CV_CALL( buffer = (uchar*)cvAlloc( buf_size ));
1421         }
1422 
1423         CV_CALL( cvInitMatHeader( &tmat, size.height, size.width, worktype, buffer ));
1424         if( type == worktype )
1425         {
1426             CV_CALL( cvCopy( src, &tmat ));
1427         }
1428         else
1429             CV_CALL( cvConvert( src, &tmat ));
1430 
1431         if( src2->data.ptr != dst->data.ptr )
1432         {
1433             CV_CALL( cvCopy( src2, dst ));
1434         }
1435 
1436         decomp_func = (CvLUDecompFunc)(lu_decomp_tab.fn_2d[CV_MAT_DEPTH(type)-CV_32F]);
1437         back_func = (CvLUBackFunc)(lu_back_tab.fn_2d[CV_MAT_DEPTH(type)-CV_32F]);
1438         assert( decomp_func && back_func );
1439 
1440         IPPI_CALL( decomp_func( tmat.data.db, tmat.step, size,
1441                                 dst->data.ptr, dst->step, dstsize, &d ));
1442 
1443         if( d != 0 )
1444         {
1445             IPPI_CALL( back_func( tmat.data.db, tmat.step, size,
1446                                   dst->data.ptr, dst->step, dstsize ));
1447         }
1448         else
1449             result = 0;
1450     }
1451 
1452     if( !result )
1453         CV_CALL( cvSetZero( dst ));
1454 
1455     __END__;
1456 
1457     if( buffer && !local_alloc )
1458         cvFree( &buffer );
1459 
1460     if( u || v || w )
1461     {
1462         cvReleaseMat( &u );
1463         cvReleaseMat( &v );
1464         cvReleaseMat( &w );
1465     }
1466 
1467     return result;
1468 }
1469 
1470 
1471 
1472 /****************************************************************************************\
1473 *                               3D vector cross-product                                  *
1474 \****************************************************************************************/
1475 
1476 CV_IMPL void
cvCrossProduct(const CvArr * srcAarr,const CvArr * srcBarr,CvArr * dstarr)1477 cvCrossProduct( const CvArr* srcAarr, const CvArr* srcBarr, CvArr* dstarr )
1478 {
1479     CV_FUNCNAME( "cvCrossProduct" );
1480 
1481     __BEGIN__;
1482 
1483     CvMat stubA, *srcA = (CvMat*)srcAarr;
1484     CvMat stubB, *srcB = (CvMat*)srcBarr;
1485     CvMat dstub, *dst = (CvMat*)dstarr;
1486     int type;
1487 
1488     if( !CV_IS_MAT(srcA))
1489         CV_CALL( srcA = cvGetMat( srcA, &stubA ));
1490 
1491     type = CV_MAT_TYPE( srcA->type );
1492 
1493     if( srcA->width*srcA->height*CV_MAT_CN(type) != 3 )
1494         CV_ERROR( CV_StsBadArg, "All the input arrays must be continuous 3-vectors" );
1495 
1496     if( !srcB || !dst )
1497         CV_ERROR( CV_StsNullPtr, "" );
1498 
1499     if( (srcA->type & ~CV_MAT_CONT_FLAG) == (srcB->type & ~CV_MAT_CONT_FLAG) &&
1500         (srcA->type & ~CV_MAT_CONT_FLAG) == (dst->type & ~CV_MAT_CONT_FLAG) )
1501     {
1502         if( !srcB->data.ptr || !dst->data.ptr )
1503             CV_ERROR( CV_StsNullPtr, "" );
1504     }
1505     else
1506     {
1507         if( !CV_IS_MAT(srcB))
1508             CV_CALL( srcB = cvGetMat( srcB, &stubB ));
1509 
1510         if( !CV_IS_MAT(dst))
1511             CV_CALL( dst = cvGetMat( dst, &dstub ));
1512 
1513         if( !CV_ARE_TYPES_EQ( srcA, srcB ) ||
1514             !CV_ARE_TYPES_EQ( srcB, dst ))
1515             CV_ERROR( CV_StsUnmatchedFormats, "" );
1516     }
1517 
1518     if( !CV_ARE_SIZES_EQ( srcA, srcB ) || !CV_ARE_SIZES_EQ( srcB, dst ))
1519         CV_ERROR( CV_StsUnmatchedSizes, "" );
1520 
1521     if( CV_MAT_DEPTH(type) == CV_32F )
1522     {
1523         float* dstdata = (float*)(dst->data.ptr);
1524         const float* src1data = (float*)(srcA->data.ptr);
1525         const float* src2data = (float*)(srcB->data.ptr);
1526 
1527         if( CV_IS_MAT_CONT(srcA->type & srcB->type & dst->type) )
1528         {
1529             dstdata[2] = src1data[0] * src2data[1] - src1data[1] * src2data[0];
1530             dstdata[0] = src1data[1] * src2data[2] - src1data[2] * src2data[1];
1531             dstdata[1] = src1data[2] * src2data[0] - src1data[0] * src2data[2];
1532         }
1533         else
1534         {
1535             int step1 = srcA->step ? srcA->step/sizeof(src1data[0]) : 1;
1536             int step2 = srcB->step ? srcB->step/sizeof(src1data[0]) : 1;
1537             int step = dst->step ? dst->step/sizeof(src1data[0]) : 1;
1538 
1539             dstdata[2*step] = src1data[0] * src2data[step2] - src1data[step1] * src2data[0];
1540             dstdata[0] = src1data[step1] * src2data[step2*2] - src1data[step1*2] * src2data[step2];
1541             dstdata[step] = src1data[step1*2] * src2data[0] - src1data[0] * src2data[step2*2];
1542         }
1543     }
1544     else if( CV_MAT_DEPTH(type) == CV_64F )
1545     {
1546         double* dstdata = (double*)(dst->data.ptr);
1547         const double* src1data = (double*)(srcA->data.ptr);
1548         const double* src2data = (double*)(srcB->data.ptr);
1549 
1550         if( CV_IS_MAT_CONT(srcA->type & srcB->type & dst->type) )
1551         {
1552             dstdata[2] = src1data[0] * src2data[1] - src1data[1] * src2data[0];
1553             dstdata[0] = src1data[1] * src2data[2] - src1data[2] * src2data[1];
1554             dstdata[1] = src1data[2] * src2data[0] - src1data[0] * src2data[2];
1555         }
1556         else
1557         {
1558             int step1 = srcA->step ? srcA->step/sizeof(src1data[0]) : 1;
1559             int step2 = srcB->step ? srcB->step/sizeof(src1data[0]) : 1;
1560             int step = dst->step ? dst->step/sizeof(src1data[0]) : 1;
1561 
1562             dstdata[2*step] = src1data[0] * src2data[step2] - src1data[step1] * src2data[0];
1563             dstdata[0] = src1data[step1] * src2data[step2*2] - src1data[step1*2] * src2data[step2];
1564             dstdata[step] = src1data[step1*2] * src2data[0] - src1data[0] * src2data[step2*2];
1565         }
1566     }
1567     else
1568     {
1569         CV_ERROR( CV_StsUnsupportedFormat, "" );
1570     }
1571 
1572     __END__;
1573 }
1574 
1575 
1576 CV_IMPL void
cvCalcPCA(const CvArr * data_arr,CvArr * avg_arr,CvArr * eigenvals,CvArr * eigenvects,int flags)1577 cvCalcPCA( const CvArr* data_arr, CvArr* avg_arr, CvArr* eigenvals, CvArr* eigenvects, int flags )
1578 {
1579     CvMat* tmp_avg = 0;
1580     CvMat* tmp_avg_r = 0;
1581     CvMat* tmp_cov = 0;
1582     CvMat* tmp_evals = 0;
1583     CvMat* tmp_evects = 0;
1584     CvMat* tmp_evects2 = 0;
1585     CvMat* tmp_data = 0;
1586 
1587     CV_FUNCNAME( "cvCalcPCA" );
1588 
1589     __BEGIN__;
1590 
1591     CvMat stub, *data = (CvMat*)data_arr;
1592     CvMat astub, *avg = (CvMat*)avg_arr;
1593     CvMat evalstub, *evals = (CvMat*)eigenvals;
1594     CvMat evectstub, *evects = (CvMat*)eigenvects;
1595     int covar_flags = CV_COVAR_SCALE;
1596     int i, len, in_count, count, out_count;
1597 
1598     if( !CV_IS_MAT(data) )
1599         CV_CALL( data = cvGetMat( data, &stub ));
1600 
1601     if( !CV_IS_MAT(avg) )
1602         CV_CALL( avg = cvGetMat( avg, &astub ));
1603 
1604     if( !CV_IS_MAT(evals) )
1605         CV_CALL( evals = cvGetMat( evals, &evalstub ));
1606 
1607     if( !CV_IS_MAT(evects) )
1608         CV_CALL( evects = cvGetMat( evects, &evectstub ));
1609 
1610     if( CV_MAT_CN(data->type) != 1 || CV_MAT_CN(avg->type) != 1 ||
1611         CV_MAT_CN(evals->type) != 1 || CV_MAT_CN(evects->type) != 1 )
1612         CV_ERROR( CV_StsUnsupportedFormat, "All the input and output arrays must be 1-channel" );
1613 
1614     if( CV_MAT_DEPTH(avg->type) < CV_32F || !CV_ARE_DEPTHS_EQ(avg, evals) ||
1615         !CV_ARE_DEPTHS_EQ(avg, evects) )
1616         CV_ERROR( CV_StsUnsupportedFormat, "All the output arrays must have the same type, 32fC1 or 64fC1" );
1617 
1618     if( flags & CV_PCA_DATA_AS_COL )
1619     {
1620         len = data->rows;
1621         in_count = data->cols;
1622         covar_flags |= CV_COVAR_COLS;
1623 
1624         if( avg->cols != 1 || avg->rows != len )
1625             CV_ERROR( CV_StsBadSize,
1626             "The mean (average) vector should be data->rows x 1 when CV_PCA_DATA_AS_COL is used" );
1627 
1628         CV_CALL( tmp_avg = cvCreateMat( len, 1, CV_64F ));
1629     }
1630     else
1631     {
1632         len = data->cols;
1633         in_count = data->rows;
1634         covar_flags |= CV_COVAR_ROWS;
1635 
1636         if( avg->rows != 1 || avg->cols != len )
1637             CV_ERROR( CV_StsBadSize,
1638             "The mean (average) vector should be 1 x data->cols when CV_PCA_DATA_AS_ROW is used" );
1639 
1640         CV_CALL( tmp_avg = cvCreateMat( 1, len, CV_64F ));
1641     }
1642 
1643     count = MIN(len, in_count);
1644     out_count = evals->cols + evals->rows - 1;
1645 
1646     if( (evals->cols != 1 && evals->rows != 1) || out_count > count )
1647         CV_ERROR( CV_StsBadSize,
1648         "The array of eigenvalues must be 1d vector containing "
1649         "no more than min(data->rows,data->cols) elements" );
1650 
1651     if( evects->cols != len || evects->rows != out_count )
1652         CV_ERROR( CV_StsBadSize,
1653         "The matrix of eigenvalues must have the same number of columns as the input vector length "
1654         "and the same number of rows as the number of eigenvalues" );
1655 
1656     // "scrambled" way to compute PCA (when cols(A)>rows(A)):
1657     // B = A'A; B*x=b*x; C = AA'; C*y=c*y -> AA'*y=c*y -> A'A*(A'*y)=c*(A'*y) -> c = b, x=A'*y
1658     if( len <= in_count )
1659         covar_flags |= CV_COVAR_NORMAL;
1660 
1661     if( flags & CV_PCA_USE_AVG ){
1662         covar_flags |= CV_COVAR_USE_AVG;
1663 		CV_CALL( cvConvert( avg, tmp_avg ) );
1664 	}
1665 
1666     CV_CALL( tmp_cov = cvCreateMat( count, count, CV_64F ));
1667     CV_CALL( tmp_evals = cvCreateMat( 1, count, CV_64F ));
1668     CV_CALL( tmp_evects = cvCreateMat( count, count, CV_64F ));
1669 
1670     CV_CALL( cvCalcCovarMatrix( &data_arr, 0, tmp_cov, tmp_avg, covar_flags ));
1671     CV_CALL( cvSVD( tmp_cov, tmp_evals, tmp_evects, 0, CV_SVD_MODIFY_A + CV_SVD_U_T ));
1672     tmp_evects->rows = out_count;
1673     tmp_evals->cols = out_count;
1674     cvZero( evects );
1675     cvZero( evals );
1676 
1677     if( covar_flags & CV_COVAR_NORMAL )
1678     {
1679         CV_CALL( cvConvert( tmp_evects, evects ));
1680     }
1681     else
1682     {
1683         // CV_PCA_DATA_AS_ROW: cols(A)>rows(A). x=A'*y -> x'=y'*A
1684         // CV_PCA_DATA_AS_COL: rows(A)>cols(A). x=A''*y -> x'=y'*A'
1685         int block_count = 0;
1686 
1687         CV_CALL( tmp_data = cvCreateMat( count, count, CV_64F ));
1688         CV_CALL( tmp_avg_r = cvCreateMat( count, count, CV_64F ));
1689         CV_CALL( tmp_evects2 = cvCreateMat( out_count, count, CV_64F ));
1690 
1691         for( i = 0; i < len; i += block_count )
1692         {
1693             CvMat data_part, tdata_part, part, dst_part, avg_part, tmp_avg_part;
1694             int gemm_flags;
1695 
1696             block_count = MIN( count, len - i );
1697 
1698             if( flags & CV_PCA_DATA_AS_COL )
1699             {
1700                 cvGetRows( data, &data_part, i, i + block_count );
1701                 cvGetRows( tmp_data, &tdata_part, 0, block_count );
1702                 cvGetRows( tmp_avg, &avg_part, i, i + block_count );
1703                 cvGetRows( tmp_avg_r, &tmp_avg_part, 0, block_count );
1704                 gemm_flags = CV_GEMM_B_T;
1705             }
1706             else
1707             {
1708                 cvGetCols( data, &data_part, i, i + block_count );
1709                 cvGetCols( tmp_data, &tdata_part, 0, block_count );
1710                 cvGetCols( tmp_avg, &avg_part, i, i + block_count );
1711                 cvGetCols( tmp_avg_r, &tmp_avg_part, 0, block_count );
1712                 gemm_flags = 0;
1713             }
1714 
1715             cvGetCols( tmp_evects2, &part, 0, block_count );
1716             cvGetCols( evects, &dst_part, i, i + block_count );
1717 
1718             cvConvert( &data_part, &tdata_part );
1719             cvRepeat( &avg_part, &tmp_avg_part );
1720             cvSub( &tdata_part, &tmp_avg_part, &tdata_part );
1721             cvGEMM( tmp_evects, &tdata_part, 1, 0, 0, &part, gemm_flags );
1722             cvConvert( &part, &dst_part );
1723         }
1724 
1725         // normalize eigenvectors
1726         for( i = 0; i < out_count; i++ )
1727         {
1728             CvMat ei;
1729             cvGetRow( evects, &ei, i );
1730 			cvNormalize( &ei, &ei );
1731         }
1732     }
1733 
1734     if( tmp_evals->rows != evals->rows )
1735         cvReshape( tmp_evals, tmp_evals, 1, evals->rows );
1736     cvConvert( tmp_evals, evals );
1737     cvConvert( tmp_avg, avg );
1738 
1739     __END__;
1740 
1741     cvReleaseMat( &tmp_avg );
1742     cvReleaseMat( &tmp_avg_r );
1743     cvReleaseMat( &tmp_cov );
1744     cvReleaseMat( &tmp_evals );
1745     cvReleaseMat( &tmp_evects );
1746     cvReleaseMat( &tmp_evects2 );
1747     cvReleaseMat( &tmp_data );
1748 }
1749 
1750 
1751 CV_IMPL void
cvProjectPCA(const CvArr * data_arr,const CvArr * avg_arr,const CvArr * eigenvects,CvArr * result_arr)1752 cvProjectPCA( const CvArr* data_arr, const CvArr* avg_arr,
1753               const CvArr* eigenvects, CvArr* result_arr )
1754 {
1755     uchar* buffer = 0;
1756     int local_alloc = 0;
1757 
1758     CV_FUNCNAME( "cvProjectPCA" );
1759 
1760     __BEGIN__;
1761 
1762     CvMat stub, *data = (CvMat*)data_arr;
1763     CvMat astub, *avg = (CvMat*)avg_arr;
1764     CvMat evectstub, *evects = (CvMat*)eigenvects;
1765     CvMat rstub, *result = (CvMat*)result_arr;
1766     CvMat avg_repeated;
1767     int i, len, in_count;
1768     int gemm_flags, as_cols, convert_data;
1769     int block_count0, block_count, buf_size, elem_size;
1770     uchar* tmp_data_ptr;
1771 
1772     if( !CV_IS_MAT(data) )
1773         CV_CALL( data = cvGetMat( data, &stub ));
1774 
1775     if( !CV_IS_MAT(avg) )
1776         CV_CALL( avg = cvGetMat( avg, &astub ));
1777 
1778     if( !CV_IS_MAT(evects) )
1779         CV_CALL( evects = cvGetMat( evects, &evectstub ));
1780 
1781     if( !CV_IS_MAT(result) )
1782         CV_CALL( result = cvGetMat( result, &rstub ));
1783 
1784     if( CV_MAT_CN(data->type) != 1 || CV_MAT_CN(avg->type) != 1 )
1785         CV_ERROR( CV_StsUnsupportedFormat, "All the input and output arrays must be 1-channel" );
1786 
1787     if( (CV_MAT_TYPE(avg->type) != CV_32FC1 && CV_MAT_TYPE(avg->type) != CV_64FC1) ||
1788         !CV_ARE_TYPES_EQ(avg, evects) || !CV_ARE_TYPES_EQ(avg, result) )
1789         CV_ERROR( CV_StsUnsupportedFormat,
1790         "All the input and output arrays (except for data) must have the same type, 32fC1 or 64fC1" );
1791 
1792     if( (avg->cols != 1 || avg->rows != data->rows) &&
1793         (avg->rows != 1 || avg->cols != data->cols) )
1794         CV_ERROR( CV_StsBadSize,
1795         "The mean (average) vector should be either 1 x data->cols or data->rows x 1" );
1796 
1797     if( avg->cols == 1 )
1798     {
1799         len = data->rows;
1800         in_count = data->cols;
1801 
1802         gemm_flags = CV_GEMM_A_T + CV_GEMM_B_T;
1803         as_cols = 1;
1804     }
1805     else
1806     {
1807         len = data->cols;
1808         in_count = data->rows;
1809 
1810         gemm_flags = CV_GEMM_B_T;
1811         as_cols = 0;
1812     }
1813 
1814     if( evects->cols != len )
1815         CV_ERROR( CV_StsUnmatchedSizes,
1816         "Eigenvectors must be stored as rows and be of the same size as input vectors" );
1817 
1818     if( result->cols > evects->rows )
1819         CV_ERROR( CV_StsOutOfRange,
1820         "The output matrix of coefficients must have the number of columns "
1821         "less than or equal to the number of eigenvectors (number of rows in eigenvectors matrix)" );
1822 
1823     evects = cvGetRows( evects, &evectstub, 0, result->cols );
1824 
1825     block_count0 = (1 << 16)/len;
1826     block_count0 = MAX( block_count0, 4 );
1827     block_count0 = MIN( block_count0, in_count );
1828     elem_size = CV_ELEM_SIZE(avg->type);
1829     convert_data = CV_MAT_DEPTH(data->type) < CV_MAT_DEPTH(avg->type);
1830 
1831     buf_size = block_count0*len*((block_count0 > 1) + 1)*elem_size;
1832 
1833     if( buf_size < CV_MAX_LOCAL_SIZE )
1834     {
1835         buffer = (uchar*)cvStackAlloc( buf_size );
1836         local_alloc = 1;
1837     }
1838     else
1839         CV_CALL( buffer = (uchar*)cvAlloc( buf_size ));
1840 
1841     tmp_data_ptr = buffer;
1842     if( block_count0 > 1 )
1843     {
1844         avg_repeated = cvMat( as_cols ? len : block_count0,
1845                               as_cols ? block_count0 : len, avg->type, buffer );
1846         cvRepeat( avg, &avg_repeated );
1847         tmp_data_ptr += block_count0*len*elem_size;
1848     }
1849     else
1850         avg_repeated = *avg;
1851 
1852     for( i = 0; i < in_count; i += block_count )
1853     {
1854         CvMat data_part, norm_data, avg_part, *src = &data_part, out_part;
1855 
1856         block_count = MIN( block_count0, in_count - i );
1857         if( as_cols )
1858         {
1859             cvGetCols( data, &data_part, i, i + block_count );
1860             cvGetCols( &avg_repeated, &avg_part, 0, block_count );
1861             norm_data = cvMat( len, block_count, avg->type, tmp_data_ptr );
1862         }
1863         else
1864         {
1865             cvGetRows( data, &data_part, i, i + block_count );
1866             cvGetRows( &avg_repeated, &avg_part, 0, block_count );
1867             norm_data = cvMat( block_count, len, avg->type, tmp_data_ptr );
1868         }
1869 
1870         if( convert_data )
1871         {
1872             cvConvert( src, &norm_data );
1873             src = &norm_data;
1874         }
1875 
1876         cvSub( src, &avg_part, &norm_data );
1877 
1878         cvGetRows( result, &out_part, i, i + block_count );
1879         cvGEMM( &norm_data, evects, 1, 0, 0, &out_part, gemm_flags );
1880     }
1881 
1882     __END__;
1883 
1884     if( !local_alloc )
1885         cvFree( &buffer );
1886 }
1887 
1888 
1889 CV_IMPL void
cvBackProjectPCA(const CvArr * proj_arr,const CvArr * avg_arr,const CvArr * eigenvects,CvArr * result_arr)1890 cvBackProjectPCA( const CvArr* proj_arr, const CvArr* avg_arr,
1891                   const CvArr* eigenvects, CvArr* result_arr )
1892 {
1893     uchar* buffer = 0;
1894     int local_alloc = 0;
1895 
1896     CV_FUNCNAME( "cvProjectPCA" );
1897 
1898     __BEGIN__;
1899 
1900     CvMat pstub, *data = (CvMat*)proj_arr;
1901     CvMat astub, *avg = (CvMat*)avg_arr;
1902     CvMat evectstub, *evects = (CvMat*)eigenvects;
1903     CvMat rstub, *result = (CvMat*)result_arr;
1904     CvMat avg_repeated;
1905     int i, len, in_count, as_cols;
1906     int block_count0, block_count, buf_size, elem_size;
1907 
1908     if( !CV_IS_MAT(data) )
1909         CV_CALL( data = cvGetMat( data, &pstub ));
1910 
1911     if( !CV_IS_MAT(avg) )
1912         CV_CALL( avg = cvGetMat( avg, &astub ));
1913 
1914     if( !CV_IS_MAT(evects) )
1915         CV_CALL( evects = cvGetMat( evects, &evectstub ));
1916 
1917     if( !CV_IS_MAT(result) )
1918         CV_CALL( result = cvGetMat( result, &rstub ));
1919 
1920     if( (CV_MAT_TYPE(avg->type) != CV_32FC1 && CV_MAT_TYPE(avg->type) != CV_64FC1) ||
1921         !CV_ARE_TYPES_EQ(avg, data) || !CV_ARE_TYPES_EQ(avg, evects) || !CV_ARE_TYPES_EQ(avg, result) )
1922         CV_ERROR( CV_StsUnsupportedFormat,
1923         "All the input and output arrays must have the same type, 32fC1 or 64fC1" );
1924 
1925     if( (avg->cols != 1 || avg->rows != result->rows) &&
1926         (avg->rows != 1 || avg->cols != result->cols) )
1927         CV_ERROR( CV_StsBadSize,
1928         "The mean (average) vector should be either 1 x result->cols or result->rows x 1" );
1929 
1930     if( avg->cols == 1 )
1931     {
1932         len = result->rows;
1933         in_count = result->cols;
1934         as_cols = 1;
1935     }
1936     else
1937     {
1938         len = result->cols;
1939         in_count = result->rows;
1940         as_cols = 0;
1941     }
1942 
1943     if( evects->cols != len )
1944         CV_ERROR( CV_StsUnmatchedSizes,
1945         "Eigenvectors must be stored as rows and be of the same size as the output vectors" );
1946 
1947     if( data->cols > evects->rows )
1948         CV_ERROR( CV_StsOutOfRange,
1949         "The input matrix of coefficients must have the number of columns "
1950         "less than or equal to the number of eigenvectors (number of rows in eigenvectors matrix)" );
1951 
1952     evects = cvGetRows( evects, &evectstub, 0, data->cols );
1953 
1954     block_count0 = (1 << 16)/len;
1955     block_count0 = MAX( block_count0, 4 );
1956     block_count0 = MIN( block_count0, in_count );
1957     elem_size = CV_ELEM_SIZE(avg->type);
1958 
1959     buf_size = block_count0*len*(block_count0 > 1)*elem_size;
1960 
1961     if( buf_size < CV_MAX_LOCAL_SIZE )
1962     {
1963         buffer = (uchar*)cvStackAlloc( MAX(buf_size,16) );
1964         local_alloc = 1;
1965     }
1966     else
1967         CV_CALL( buffer = (uchar*)cvAlloc( buf_size ));
1968 
1969     if( block_count0 > 1 )
1970     {
1971         avg_repeated = cvMat( as_cols ? len : block_count0,
1972                               as_cols ? block_count0 : len, avg->type, buffer );
1973         cvRepeat( avg, &avg_repeated );
1974     }
1975     else
1976         avg_repeated = *avg;
1977 
1978     for( i = 0; i < in_count; i += block_count )
1979     {
1980         CvMat data_part, avg_part, out_part;
1981 
1982         block_count = MIN( block_count0, in_count - i );
1983         cvGetRows( data, &data_part, i, i + block_count );
1984 
1985         if( as_cols )
1986         {
1987             cvGetCols( result, &out_part, i, i + block_count );
1988             cvGetCols( &avg_repeated, &avg_part, 0, block_count );
1989             cvGEMM( evects, &data_part, 1, &avg_part, 1, &out_part, CV_GEMM_A_T + CV_GEMM_B_T );
1990         }
1991         else
1992         {
1993             cvGetRows( result, &out_part, i, i + block_count );
1994             cvGetRows( &avg_repeated, &avg_part, 0, block_count );
1995             cvGEMM( &data_part, evects, 1, &avg_part, 1, &out_part, 0 );
1996         }
1997     }
1998 
1999     __END__;
2000 
2001     if( !local_alloc )
2002         cvFree( &buffer );
2003 }
2004 
2005 
2006 /* End of file. */
2007