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
44void
45icvCrossCorr( 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
309icvCrossCorrValid_Norm_8u32f_C1R_t  icvCrossCorrValid_Norm_8u32f_C1R_p = 0;
310icvCrossCorrValid_NormLevel_8u32f_C1R_t  icvCrossCorrValid_NormLevel_8u32f_C1R_p = 0;
311icvSqrDistanceValid_Norm_8u32f_C1R_t  icvSqrDistanceValid_Norm_8u32f_C1R_p = 0;
312icvCrossCorrValid_Norm_32f_C1R_t  icvCrossCorrValid_Norm_32f_C1R_p = 0;
313icvCrossCorrValid_NormLevel_32f_C1R_t  icvCrossCorrValid_NormLevel_32f_C1R_p = 0;
314icvSqrDistanceValid_Norm_32f_C1R_t  icvSqrDistanceValid_Norm_32f_C1R_p = 0;
315
316typedef 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
323CV_IMPL void
324cvMatchTemplate( 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