• 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 "_cv.h"
43 
44 void
icvCrossCorr(const CvArr * _img,const CvArr * _templ,CvArr * _corr,CvPoint anchor)45 icvCrossCorr( const CvArr* _img, const CvArr* _templ, CvArr* _corr, CvPoint anchor )
46 {
47     const double block_scale = 4.5;
48     const int min_block_size = 256;
49     CvMat* dft_img[CV_MAX_THREADS] = {0};
50     CvMat* dft_templ = 0;
51     void* buf[CV_MAX_THREADS] = {0};
52     int k, num_threads = 0;
53 
54     CV_FUNCNAME( "icvCrossCorr" );
55 
56     __BEGIN__;
57 
58     CvMat istub, *img = (CvMat*)_img;
59     CvMat tstub, *templ = (CvMat*)_templ;
60     CvMat cstub, *corr = (CvMat*)_corr;
61     CvSize dftsize, blocksize;
62     int depth, templ_depth, corr_depth, max_depth = CV_32F,
63         cn, templ_cn, corr_cn, buf_size = 0,
64         tile_count_x, tile_count_y, tile_count;
65 
66     CV_CALL( img = cvGetMat( img, &istub ));
67     CV_CALL( templ = cvGetMat( templ, &tstub ));
68     CV_CALL( corr = cvGetMat( corr, &cstub ));
69 
70     if( CV_MAT_DEPTH( img->type ) != CV_8U &&
71         CV_MAT_DEPTH( img->type ) != CV_16U &&
72         CV_MAT_DEPTH( img->type ) != CV_32F )
73         CV_ERROR( CV_StsUnsupportedFormat,
74         "The function supports only 8u, 16u and 32f data types" );
75 
76     if( !CV_ARE_DEPTHS_EQ( img, templ ) && CV_MAT_DEPTH( templ->type ) != CV_32F )
77         CV_ERROR( CV_StsUnsupportedFormat,
78         "Template (kernel) must be of the same depth as the input image, or be 32f" );
79 
80     if( !CV_ARE_DEPTHS_EQ( img, corr ) && CV_MAT_DEPTH( corr->type ) != CV_32F &&
81         CV_MAT_DEPTH( corr->type ) != CV_64F )
82         CV_ERROR( CV_StsUnsupportedFormat,
83         "The output image must have the same depth as the input image, or be 32f/64f" );
84 
85     if( (!CV_ARE_CNS_EQ( img, corr ) || CV_MAT_CN(templ->type) > 1) &&
86         (CV_MAT_CN( corr->type ) > 1 || !CV_ARE_CNS_EQ( img, templ)) )
87         CV_ERROR( CV_StsUnsupportedFormat,
88         "The output must have the same number of channels as the input (when the template has 1 channel), "
89         "or the output must have 1 channel when the input and the template have the same number of channels" );
90 
91     depth = CV_MAT_DEPTH(img->type);
92     cn = CV_MAT_CN(img->type);
93     templ_depth = CV_MAT_DEPTH(templ->type);
94     templ_cn = CV_MAT_CN(templ->type);
95     corr_depth = CV_MAT_DEPTH(corr->type);
96     corr_cn = CV_MAT_CN(corr->type);
97     max_depth = MAX( max_depth, templ_depth );
98     max_depth = MAX( max_depth, depth );
99     max_depth = MAX( max_depth, corr_depth );
100     if( depth > CV_8U )
101         max_depth = CV_64F;
102 
103     if( img->cols < templ->cols || img->rows < templ->rows )
104         CV_ERROR( CV_StsUnmatchedSizes,
105         "Such a combination of image and template/filter size is not supported" );
106 
107     if( corr->rows > img->rows + templ->rows - 1 ||
108         corr->cols > img->cols + templ->cols - 1 )
109         CV_ERROR( CV_StsUnmatchedSizes,
110         "output image should not be greater than (W + w - 1)x(H + h - 1)" );
111 
112     blocksize.width = cvRound(templ->cols*block_scale);
113     blocksize.width = MAX( blocksize.width, min_block_size - templ->cols + 1 );
114     blocksize.width = MIN( blocksize.width, corr->cols );
115     blocksize.height = cvRound(templ->rows*block_scale);
116     blocksize.height = MAX( blocksize.height, min_block_size - templ->rows + 1 );
117     blocksize.height = MIN( blocksize.height, corr->rows );
118 
119     dftsize.width = cvGetOptimalDFTSize(blocksize.width + templ->cols - 1);
120     if( dftsize.width == 1 )
121         dftsize.width = 2;
122     dftsize.height = cvGetOptimalDFTSize(blocksize.height + templ->rows - 1);
123     if( dftsize.width <= 0 || dftsize.height <= 0 )
124         CV_ERROR( CV_StsOutOfRange, "the input arrays are too big" );
125 
126     // recompute block size
127     blocksize.width = dftsize.width - templ->cols + 1;
128     blocksize.width = MIN( blocksize.width, corr->cols );
129     blocksize.height = dftsize.height - templ->rows + 1;
130     blocksize.height = MIN( blocksize.height, corr->rows );
131 
132     CV_CALL( dft_templ = cvCreateMat( dftsize.height*templ_cn, dftsize.width, max_depth ));
133 
134     num_threads = cvGetNumThreads();
135 
136     for( k = 0; k < num_threads; k++ )
137         CV_CALL( dft_img[k] = cvCreateMat( dftsize.height, dftsize.width, max_depth ));
138 
139     if( templ_cn > 1 && templ_depth != max_depth )
140         buf_size = templ->cols*templ->rows*CV_ELEM_SIZE(templ_depth);
141 
142     if( cn > 1 && depth != max_depth )
143         buf_size = MAX( buf_size, (blocksize.width + templ->cols - 1)*
144             (blocksize.height + templ->rows - 1)*CV_ELEM_SIZE(depth));
145 
146     if( (corr_cn > 1 || cn > 1) && corr_depth != max_depth )
147         buf_size = MAX( buf_size, blocksize.width*blocksize.height*CV_ELEM_SIZE(corr_depth));
148 
149     if( buf_size > 0 )
150     {
151         for( k = 0; k < num_threads; k++ )
152             CV_CALL( buf[k] = cvAlloc(buf_size) );
153     }
154 
155     // compute DFT of each template plane
156     for( k = 0; k < templ_cn; k++ )
157     {
158         CvMat dstub, *src, *dst, temp;
159         CvMat* planes[] = { 0, 0, 0, 0 };
160         int yofs = k*dftsize.height;
161 
162         src = templ;
163         dst = cvGetSubRect( dft_templ, &dstub, cvRect(0,yofs,templ->cols,templ->rows));
164 
165         if( templ_cn > 1 )
166         {
167             planes[k] = templ_depth == max_depth ? dst :
168                 cvInitMatHeader( &temp, templ->rows, templ->cols, templ_depth, buf[0] );
169             cvSplit( templ, planes[0], planes[1], planes[2], planes[3] );
170             src = planes[k];
171             planes[k] = 0;
172         }
173 
174         if( dst != src )
175             cvConvert( src, dst );
176 
177         if( dft_templ->cols > templ->cols )
178         {
179             cvGetSubRect( dft_templ, dst, cvRect(templ->cols, yofs,
180                           dft_templ->cols - templ->cols, templ->rows) );
181             cvZero( dst );
182         }
183         cvGetSubRect( dft_templ, dst, cvRect(0,yofs,dftsize.width,dftsize.height) );
184         cvDFT( dst, dst, CV_DXT_FORWARD + CV_DXT_SCALE, templ->rows );
185     }
186 
187     tile_count_x = (corr->cols + blocksize.width - 1)/blocksize.width;
188     tile_count_y = (corr->rows + blocksize.height - 1)/blocksize.height;
189     tile_count = tile_count_x*tile_count_y;
190 
191     {
192 #ifdef _OPENMP
193     #pragma omp parallel for num_threads(num_threads) schedule(dynamic)
194 #endif
195     // calculate correlation by blocks
196     for( k = 0; k < tile_count; k++ )
197     {
198         int thread_idx = cvGetThreadNum();
199         int x = (k%tile_count_x)*blocksize.width;
200         int y = (k/tile_count_x)*blocksize.height;
201         int i, yofs;
202         CvMat sstub, dstub, *src, *dst, temp;
203         CvMat* planes[] = { 0, 0, 0, 0 };
204         CvMat* _dft_img = dft_img[thread_idx];
205         void* _buf = buf[thread_idx];
206         CvSize csz = { blocksize.width, blocksize.height }, isz;
207         int x0 = x - anchor.x, y0 = y - anchor.y;
208         int x1 = MAX( 0, x0 ), y1 = MAX( 0, y0 ), x2, y2;
209         csz.width = MIN( csz.width, corr->cols - x );
210         csz.height = MIN( csz.height, corr->rows - y );
211         isz.width = csz.width + templ->cols - 1;
212         isz.height = csz.height + templ->rows - 1;
213         x2 = MIN( img->cols, x0 + isz.width );
214         y2 = MIN( img->rows, y0 + isz.height );
215 
216         for( i = 0; i < cn; i++ )
217         {
218             CvMat dstub1, *dst1;
219             yofs = i*dftsize.height;
220 
221             src = cvGetSubRect( img, &sstub, cvRect(x1,y1,x2-x1,y2-y1) );
222             dst = cvGetSubRect( _dft_img, &dstub,
223                 cvRect(0,0,isz.width,isz.height) );
224             dst1 = dst;
225 
226             if( x2 - x1 < isz.width || y2 - y1 < isz.height )
227                 dst1 = cvGetSubRect( _dft_img, &dstub1,
228                     cvRect( x1 - x0, y1 - y0, x2 - x1, y2 - y1 ));
229 
230             if( cn > 1 )
231             {
232                 planes[i] = dst1;
233                 if( depth != max_depth )
234                     planes[i] = cvInitMatHeader( &temp, y2 - y1, x2 - x1, depth, _buf );
235                 cvSplit( src, planes[0], planes[1], planes[2], planes[3] );
236                 src = planes[i];
237                 planes[i] = 0;
238             }
239 
240             if( dst1 != src )
241                 cvConvert( src, dst1 );
242 
243             if( dst != dst1 )
244                 cvCopyMakeBorder( dst1, dst, cvPoint(x1 - x0, y1 - y0), IPL_BORDER_REPLICATE );
245 
246             if( dftsize.width > isz.width )
247             {
248                 cvGetSubRect( _dft_img, dst, cvRect(isz.width, 0,
249                       dftsize.width - isz.width,dftsize.height) );
250                 cvZero( dst );
251             }
252 
253             cvDFT( _dft_img, _dft_img, CV_DXT_FORWARD, isz.height );
254             cvGetSubRect( dft_templ, dst,
255                 cvRect(0,(templ_cn>1?yofs:0),dftsize.width,dftsize.height) );
256 
257             cvMulSpectrums( _dft_img, dst, _dft_img, CV_DXT_MUL_CONJ );
258             cvDFT( _dft_img, _dft_img, CV_DXT_INVERSE, csz.height );
259 
260             src = cvGetSubRect( _dft_img, &sstub, cvRect(0,0,csz.width,csz.height) );
261             dst = cvGetSubRect( corr, &dstub, cvRect(x,y,csz.width,csz.height) );
262 
263             if( corr_cn > 1 )
264             {
265                 planes[i] = src;
266                 if( corr_depth != max_depth )
267                 {
268                     planes[i] = cvInitMatHeader( &temp, csz.height, csz.width,
269                                                  corr_depth, _buf );
270                     cvConvert( src, planes[i] );
271                 }
272                 cvMerge( planes[0], planes[1], planes[2], planes[3], dst );
273                 planes[i] = 0;
274             }
275             else
276             {
277                 if( i == 0 )
278                     cvConvert( src, dst );
279                 else
280                 {
281                     if( max_depth > corr_depth )
282                     {
283                         cvInitMatHeader( &temp, csz.height, csz.width,
284                                          corr_depth, _buf );
285                         cvConvert( src, &temp );
286                         src = &temp;
287                     }
288                     cvAcc( src, dst );
289                 }
290             }
291         }
292     }
293     }
294 
295     __END__;
296 
297     cvReleaseMat( &dft_templ );
298 
299     for( k = 0; k < num_threads; k++ )
300     {
301         cvReleaseMat( &dft_img[k] );
302         cvFree( &buf[k] );
303     }
304 }
305 
306 
307 /***************************** IPP Match Template Functions ******************************/
308 
309 icvCrossCorrValid_Norm_8u32f_C1R_t  icvCrossCorrValid_Norm_8u32f_C1R_p = 0;
310 icvCrossCorrValid_NormLevel_8u32f_C1R_t  icvCrossCorrValid_NormLevel_8u32f_C1R_p = 0;
311 icvSqrDistanceValid_Norm_8u32f_C1R_t  icvSqrDistanceValid_Norm_8u32f_C1R_p = 0;
312 icvCrossCorrValid_Norm_32f_C1R_t  icvCrossCorrValid_Norm_32f_C1R_p = 0;
313 icvCrossCorrValid_NormLevel_32f_C1R_t  icvCrossCorrValid_NormLevel_32f_C1R_p = 0;
314 icvSqrDistanceValid_Norm_32f_C1R_t  icvSqrDistanceValid_Norm_32f_C1R_p = 0;
315 
316 typedef CvStatus (CV_STDCALL * CvTemplMatchIPPFunc)
317     ( const void* img, int imgstep, CvSize imgsize,
318       const void* templ, int templstep, CvSize templsize,
319       void* result, int rstep );
320 
321 /*****************************************************************************************/
322 
323 CV_IMPL void
cvMatchTemplate(const CvArr * _img,const CvArr * _templ,CvArr * _result,int method)324 cvMatchTemplate( const CvArr* _img, const CvArr* _templ, CvArr* _result, int method )
325 {
326     CvMat* sum = 0;
327     CvMat* sqsum = 0;
328 
329     CV_FUNCNAME( "cvMatchTemplate" );
330 
331     __BEGIN__;
332 
333     int coi1 = 0, coi2 = 0;
334     int depth, cn;
335     int i, j, k;
336     CvMat stub, *img = (CvMat*)_img;
337     CvMat tstub, *templ = (CvMat*)_templ;
338     CvMat rstub, *result = (CvMat*)_result;
339     CvScalar templ_mean = cvScalarAll(0);
340     double templ_norm = 0, templ_sum2 = 0;
341 
342     int idx = 0, idx2 = 0;
343     double *p0, *p1, *p2, *p3;
344     double *q0, *q1, *q2, *q3;
345     double inv_area;
346     int sum_step, sqsum_step;
347     int num_type = method == CV_TM_CCORR || method == CV_TM_CCORR_NORMED ? 0 :
348                    method == CV_TM_CCOEFF || method == CV_TM_CCOEFF_NORMED ? 1 : 2;
349     int is_normed = method == CV_TM_CCORR_NORMED ||
350                     method == CV_TM_SQDIFF_NORMED ||
351                     method == CV_TM_CCOEFF_NORMED;
352 
353     CV_CALL( img = cvGetMat( img, &stub, &coi1 ));
354     CV_CALL( templ = cvGetMat( templ, &tstub, &coi2 ));
355     CV_CALL( result = cvGetMat( result, &rstub ));
356 
357     if( CV_MAT_DEPTH( img->type ) != CV_8U &&
358         CV_MAT_DEPTH( img->type ) != CV_32F )
359         CV_ERROR( CV_StsUnsupportedFormat,
360         "The function supports only 8u and 32f data types" );
361 
362     if( !CV_ARE_TYPES_EQ( img, templ ))
363         CV_ERROR( CV_StsUnmatchedSizes, "image and template should have the same type" );
364 
365     if( CV_MAT_TYPE( result->type ) != CV_32FC1 )
366         CV_ERROR( CV_StsUnsupportedFormat, "output image should have 32f type" );
367 
368     if( img->rows < templ->rows || img->cols < templ->cols )
369     {
370         CvMat* t;
371         CV_SWAP( img, templ, t );
372     }
373 
374     if( result->rows != img->rows - templ->rows + 1 ||
375         result->cols != img->cols - templ->cols + 1 )
376         CV_ERROR( CV_StsUnmatchedSizes, "output image should be (W - w + 1)x(H - h + 1)" );
377 
378     if( method < CV_TM_SQDIFF || method > CV_TM_CCOEFF_NORMED )
379         CV_ERROR( CV_StsBadArg, "unknown comparison method" );
380 
381     depth = CV_MAT_DEPTH(img->type);
382     cn = CV_MAT_CN(img->type);
383 
384     if( is_normed && cn == 1 && templ->rows > 8 && templ->cols > 8 &&
385         img->rows > templ->cols && img->cols > templ->cols )
386     {
387         CvTemplMatchIPPFunc ipp_func =
388             depth == CV_8U ?
389             (method == CV_TM_SQDIFF_NORMED ? (CvTemplMatchIPPFunc)icvSqrDistanceValid_Norm_8u32f_C1R_p :
390             method == CV_TM_CCORR_NORMED ? (CvTemplMatchIPPFunc)icvCrossCorrValid_Norm_8u32f_C1R_p :
391             (CvTemplMatchIPPFunc)icvCrossCorrValid_NormLevel_8u32f_C1R_p) :
392             (method == CV_TM_SQDIFF_NORMED ? (CvTemplMatchIPPFunc)icvSqrDistanceValid_Norm_32f_C1R_p :
393             method == CV_TM_CCORR_NORMED ? (CvTemplMatchIPPFunc)icvCrossCorrValid_Norm_32f_C1R_p :
394             (CvTemplMatchIPPFunc)icvCrossCorrValid_NormLevel_32f_C1R_p);
395 
396         if( ipp_func )
397         {
398             CvSize img_size = cvGetMatSize(img), templ_size = cvGetMatSize(templ);
399 
400             IPPI_CALL( ipp_func( img->data.ptr, img->step ? img->step : CV_STUB_STEP,
401                                  img_size, templ->data.ptr,
402                                  templ->step ? templ->step : CV_STUB_STEP,
403                                  templ_size, result->data.ptr,
404                                  result->step ? result->step : CV_STUB_STEP ));
405             for( i = 0; i < result->rows; i++ )
406             {
407                 float* rrow = (float*)(result->data.ptr + i*result->step);
408                 for( j = 0; j < result->cols; j++ )
409                 {
410                     if( fabs(rrow[j]) > 1. )
411                         rrow[j] = rrow[j] < 0 ? -1.f : 1.f;
412                 }
413             }
414             EXIT;
415         }
416     }
417 
418     CV_CALL( icvCrossCorr( img, templ, result ));
419 
420     if( method == CV_TM_CCORR )
421         EXIT;
422 
423     inv_area = 1./((double)templ->rows * templ->cols);
424 
425     CV_CALL( sum = cvCreateMat( img->rows + 1, img->cols + 1,
426                                 CV_MAKETYPE( CV_64F, cn )));
427     if( method == CV_TM_CCOEFF )
428     {
429         CV_CALL( cvIntegral( img, sum, 0, 0 ));
430         CV_CALL( templ_mean = cvAvg( templ ));
431         q0 = q1 = q2 = q3 = 0;
432     }
433     else
434     {
435         CvScalar _templ_sdv = cvScalarAll(0);
436         CV_CALL( sqsum = cvCreateMat( img->rows + 1, img->cols + 1,
437                                       CV_MAKETYPE( CV_64F, cn )));
438         CV_CALL( cvIntegral( img, sum, sqsum, 0 ));
439         CV_CALL( cvAvgSdv( templ, &templ_mean, &_templ_sdv ));
440 
441         templ_norm = CV_SQR(_templ_sdv.val[0]) + CV_SQR(_templ_sdv.val[1]) +
442                     CV_SQR(_templ_sdv.val[2]) + CV_SQR(_templ_sdv.val[3]);
443 
444         if( templ_norm < DBL_EPSILON && method == CV_TM_CCOEFF_NORMED )
445         {
446             cvSet( result, cvScalarAll(1.) );
447             EXIT;
448         }
449 
450         templ_sum2 = templ_norm +
451                      CV_SQR(templ_mean.val[0]) + CV_SQR(templ_mean.val[1]) +
452                      CV_SQR(templ_mean.val[2]) + CV_SQR(templ_mean.val[3]);
453 
454         if( num_type != 1 )
455         {
456             templ_mean = cvScalarAll(0);
457             templ_norm = templ_sum2;
458         }
459 
460         templ_sum2 /= inv_area;
461         templ_norm = sqrt(templ_norm);
462         templ_norm /= sqrt(inv_area); // care of accuracy here
463 
464         q0 = (double*)sqsum->data.ptr;
465         q1 = q0 + templ->cols*cn;
466         q2 = (double*)(sqsum->data.ptr + templ->rows*sqsum->step);
467         q3 = q2 + templ->cols*cn;
468     }
469 
470     p0 = (double*)sum->data.ptr;
471     p1 = p0 + templ->cols*cn;
472     p2 = (double*)(sum->data.ptr + templ->rows*sum->step);
473     p3 = p2 + templ->cols*cn;
474 
475     sum_step = sum ? sum->step / sizeof(double) : 0;
476     sqsum_step = sqsum ? sqsum->step / sizeof(double) : 0;
477 
478     for( i = 0; i < result->rows; i++ )
479     {
480         float* rrow = (float*)(result->data.ptr + i*result->step);
481         idx = i * sum_step;
482         idx2 = i * sqsum_step;
483 
484         for( j = 0; j < result->cols; j++, idx += cn, idx2 += cn )
485         {
486             double num = rrow[j], t;
487             double wnd_mean2 = 0, wnd_sum2 = 0;
488 
489             if( num_type == 1 )
490             {
491                 for( k = 0; k < cn; k++ )
492                 {
493                     t = p0[idx+k] - p1[idx+k] - p2[idx+k] + p3[idx+k];
494                     wnd_mean2 += CV_SQR(t);
495                     num -= t*templ_mean.val[k];
496                 }
497 
498                 wnd_mean2 *= inv_area;
499             }
500 
501             if( is_normed || num_type == 2 )
502             {
503                 for( k = 0; k < cn; k++ )
504                 {
505                     t = q0[idx2+k] - q1[idx2+k] - q2[idx2+k] + q3[idx2+k];
506                     wnd_sum2 += t;
507                 }
508 
509                 if( num_type == 2 )
510                     num = wnd_sum2 - 2*num + templ_sum2;
511             }
512 
513             if( is_normed )
514             {
515                 t = sqrt(MAX(wnd_sum2 - wnd_mean2,0))*templ_norm;
516                 if( t > DBL_EPSILON )
517                 {
518                     num /= t;
519                     if( fabs(num) > 1. )
520                         num = num > 0 ? 1 : -1;
521                 }
522                 else
523                     num = method != CV_TM_SQDIFF_NORMED || num < DBL_EPSILON ? 0 : 1;
524             }
525 
526             rrow[j] = (float)num;
527         }
528     }
529 
530     __END__;
531 
532     cvReleaseMat( &sum );
533     cvReleaseMat( &sqsum );
534 }
535 
536 /* End of file. */
537