1 /*M///////////////////////////////////////////////////////////////////////////////////////
\r
3 // IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
\r
5 // By downloading, copying, installing or using the software you agree to this license.
\r
6 // If you do not agree to this license, do not download, install,
\r
7 // copy or use the software.
\r
10 // License Agreement
\r
11 // For Open Source Computer Vision Library
\r
13 // Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
\r
14 // Copyright (C) 2009, Willow Garage Inc., all rights reserved.
\r
15 // Third party copyrights are property of their respective owners.
\r
17 // Redistribution and use in source and binary forms, with or without modification,
\r
18 // are permitted provided that the following conditions are met:
\r
20 // * Redistribution's of source code must retain the above copyright notice,
\r
21 // this list of conditions and the following disclaimer.
\r
23 // * Redistribution's in binary form must reproduce the above copyright notice,
\r
24 // this list of conditions and the following disclaimer in the documentation
\r
25 // and/or other materials provided with the distribution.
\r
27 // * The name of the copyright holders may not be used to endorse or promote products
\r
28 // derived from this software without specific prior written permission.
\r
30 // This software is provided by the copyright holders and contributors "as is" and
\r
31 // any express or implied warranties, including, but not limited to, the implied
\r
32 // warranties of merchantability and fitness for a particular purpose are disclaimed.
\r
33 // In no event shall the Intel Corporation or contributors be liable for any direct,
\r
34 // indirect, incidental, special, exemplary, or consequential damages
\r
35 // (including, but not limited to, procurement of substitute goods or services;
\r
36 // loss of use, data, or profits; or business interruption) however caused
\r
37 // and on any theory of liability, whether in contract, strict liability,
\r
38 // or tort (including negligence or otherwise) arising in any way out of
\r
39 // the use of this software, even if advised of the possibility of such damage.
\r
46 * This file includes the code, contributed by Simon Perreault
\r
47 * (the function icvMedianBlur_8u_O1)
\r
49 * Constant-time median filtering -- http://nomis80.org/ctmf.html
\r
50 * Copyright (C) 2006 Simon Perreault
\r
53 * Laboratoire de vision et systemes numeriques
\r
54 * Pavillon Adrien-Pouliot
\r
56 * Sainte-Foy, Quebec, Canada
\r
59 * perreaul@gel.ulaval.ca
\r
65 /****************************************************************************************\
\r
67 \****************************************************************************************/
\r
69 template<typename T, typename ST> struct RowSum : public BaseRowFilter
\r
71 RowSum( int _ksize, int _anchor )
\r
77 void operator()(const uchar* src, uchar* dst, int width, int cn)
\r
79 const T* S = (const T*)src;
\r
81 int i = 0, k, ksz_cn = ksize*cn;
\r
83 width = (width - 1)*cn;
\r
84 for( k = 0; k < cn; k++, S++, D++ )
\r
87 for( i = 0; i < ksz_cn; i += cn )
\r
90 for( i = 0; i < width; i += cn )
\r
92 s += S[i + ksz_cn] - S[i];
\r
100 template<typename ST, typename T> struct ColumnSum : public BaseColumnFilter
\r
102 ColumnSum( int _ksize, int _anchor, double _scale )
\r
110 void reset() { sumCount = 0; }
\r
112 void operator()(const uchar** src, uchar* dst, int dststep, int count, int width)
\r
116 bool haveScale = scale != 1;
\r
117 double _scale = scale;
\r
119 if( width != (int)sum.size() )
\r
126 if( sumCount == 0 )
\r
128 for( i = 0; i < width; i++ )
\r
130 for( ; sumCount < ksize - 1; sumCount++, src++ )
\r
132 const ST* Sp = (const ST*)src[0];
\r
133 for( i = 0; i <= width - 2; i += 2 )
\r
135 ST s0 = SUM[i] + Sp[i], s1 = SUM[i+1] + Sp[i+1];
\r
136 SUM[i] = s0; SUM[i+1] = s1;
\r
139 for( ; i < width; i++ )
\r
145 CV_Assert( sumCount == ksize-1 );
\r
149 for( ; count--; src++ )
\r
151 const ST* Sp = (const ST*)src[0];
\r
152 const ST* Sm = (const ST*)src[1-ksize];
\r
156 for( i = 0; i <= width - 2; i += 2 )
\r
158 ST s0 = SUM[i] + Sp[i], s1 = SUM[i+1] + Sp[i+1];
\r
159 D[i] = saturate_cast<T>(s0*_scale);
\r
160 D[i+1] = saturate_cast<T>(s1*_scale);
\r
161 s0 -= Sm[i]; s1 -= Sm[i+1];
\r
162 SUM[i] = s0; SUM[i+1] = s1;
\r
165 for( ; i < width; i++ )
\r
167 ST s0 = SUM[i] + Sp[i];
\r
168 D[i] = saturate_cast<T>(s0*_scale);
\r
169 SUM[i] = s0 - Sm[i];
\r
174 for( i = 0; i <= width - 2; i += 2 )
\r
176 ST s0 = SUM[i] + Sp[i], s1 = SUM[i+1] + Sp[i+1];
\r
177 D[i] = saturate_cast<T>(s0);
\r
178 D[i+1] = saturate_cast<T>(s1);
\r
179 s0 -= Sm[i]; s1 -= Sm[i+1];
\r
180 SUM[i] = s0; SUM[i+1] = s1;
\r
183 for( ; i < width; i++ )
\r
185 ST s0 = SUM[i] + Sp[i];
\r
186 D[i] = saturate_cast<T>(s0);
\r
187 SUM[i] = s0 - Sm[i];
\r
200 Ptr<BaseRowFilter> getRowSumFilter(int srcType, int sumType, int ksize, int anchor)
\r
202 int sdepth = CV_MAT_DEPTH(srcType), ddepth = CV_MAT_DEPTH(sumType);
\r
203 CV_Assert( CV_MAT_CN(sumType) == CV_MAT_CN(srcType) );
\r
208 if( sdepth == CV_8U && ddepth == CV_32S )
\r
209 return Ptr<BaseRowFilter>(new RowSum<uchar, int>(ksize, anchor));
\r
210 if( sdepth == CV_8U && ddepth == CV_64F )
\r
211 return Ptr<BaseRowFilter>(new RowSum<uchar, double>(ksize, anchor));
\r
212 if( sdepth == CV_16U && ddepth == CV_32S )
\r
213 return Ptr<BaseRowFilter>(new RowSum<ushort, int>(ksize, anchor));
\r
214 if( sdepth == CV_16U && ddepth == CV_64F )
\r
215 return Ptr<BaseRowFilter>(new RowSum<ushort, double>(ksize, anchor));
\r
216 if( sdepth == CV_16S && ddepth == CV_32S )
\r
217 return Ptr<BaseRowFilter>(new RowSum<short, int>(ksize, anchor));
\r
218 if( sdepth == CV_32S && ddepth == CV_32S )
\r
219 return Ptr<BaseRowFilter>(new RowSum<int, int>(ksize, anchor));
\r
220 if( sdepth == CV_16S && ddepth == CV_64F )
\r
221 return Ptr<BaseRowFilter>(new RowSum<short, double>(ksize, anchor));
\r
222 if( sdepth == CV_32F && ddepth == CV_64F )
\r
223 return Ptr<BaseRowFilter>(new RowSum<float, double>(ksize, anchor));
\r
224 if( sdepth == CV_64F && ddepth == CV_64F )
\r
225 return Ptr<BaseRowFilter>(new RowSum<double, double>(ksize, anchor));
\r
227 CV_Error_( CV_StsNotImplemented,
\r
228 ("Unsupported combination of source format (=%d), and buffer format (=%d)",
\r
229 srcType, sumType));
\r
231 return Ptr<BaseRowFilter>(0);
\r
235 Ptr<BaseColumnFilter> getColumnSumFilter(int sumType, int dstType, int ksize,
\r
236 int anchor, double scale)
\r
238 int sdepth = CV_MAT_DEPTH(sumType), ddepth = CV_MAT_DEPTH(dstType);
\r
239 CV_Assert( CV_MAT_CN(sumType) == CV_MAT_CN(dstType) );
\r
244 if( ddepth == CV_8U && sdepth == CV_32S )
\r
245 return Ptr<BaseColumnFilter>(new ColumnSum<int, uchar>(ksize, anchor, scale));
\r
246 if( ddepth == CV_8U && sdepth == CV_64F )
\r
247 return Ptr<BaseColumnFilter>(new ColumnSum<double, uchar>(ksize, anchor, scale));
\r
248 if( ddepth == CV_16U && sdepth == CV_32S )
\r
249 return Ptr<BaseColumnFilter>(new ColumnSum<int, ushort>(ksize, anchor, scale));
\r
250 if( ddepth == CV_16U && sdepth == CV_64F )
\r
251 return Ptr<BaseColumnFilter>(new ColumnSum<double, ushort>(ksize, anchor, scale));
\r
252 if( ddepth == CV_16S && sdepth == CV_32S )
\r
253 return Ptr<BaseColumnFilter>(new ColumnSum<int, short>(ksize, anchor, scale));
\r
254 if( ddepth == CV_16S && sdepth == CV_64F )
\r
255 return Ptr<BaseColumnFilter>(new ColumnSum<double, short>(ksize, anchor, scale));
\r
256 if( ddepth == CV_32S && sdepth == CV_32S )
\r
257 return Ptr<BaseColumnFilter>(new ColumnSum<int, int>(ksize, anchor, scale));
\r
258 if( ddepth == CV_32F && sdepth == CV_32S )
\r
259 return Ptr<BaseColumnFilter>(new ColumnSum<int, float>(ksize, anchor, scale));
\r
260 if( ddepth == CV_32F && sdepth == CV_64F )
\r
261 return Ptr<BaseColumnFilter>(new ColumnSum<double, float>(ksize, anchor, scale));
\r
262 if( ddepth == CV_64F && sdepth == CV_32S )
\r
263 return Ptr<BaseColumnFilter>(new ColumnSum<int, double>(ksize, anchor, scale));
\r
264 if( ddepth == CV_64F && sdepth == CV_64F )
\r
265 return Ptr<BaseColumnFilter>(new ColumnSum<double, double>(ksize, anchor, scale));
\r
267 CV_Error_( CV_StsNotImplemented,
\r
268 ("Unsupported combination of sum format (=%d), and destination format (=%d)",
\r
269 sumType, dstType));
\r
271 return Ptr<BaseColumnFilter>(0);
\r
275 Ptr<FilterEngine> createBoxFilter( int srcType, int dstType, Size ksize,
\r
276 Point anchor, bool normalize, int borderType )
\r
278 int sdepth = CV_MAT_DEPTH(srcType);
\r
279 int cn = CV_MAT_CN(srcType), sumType = CV_64F;
\r
280 if( sdepth < CV_32S && (!normalize ||
\r
281 ksize.width*ksize.height <= (sdepth == CV_8U ? (1<<23) :
\r
282 sdepth == CV_16U ? (1 << 15) : (1 << 16))) )
\r
284 sumType = CV_MAKETYPE( sumType, cn );
\r
286 Ptr<BaseRowFilter> rowFilter = getRowSumFilter(srcType, sumType, ksize.width, anchor.x );
\r
287 Ptr<BaseColumnFilter> columnFilter = getColumnSumFilter(sumType,
\r
288 dstType, ksize.height, anchor.y, normalize ? 1./(ksize.width*ksize.height) : 1);
\r
290 return Ptr<FilterEngine>(new FilterEngine(Ptr<BaseFilter>(0), rowFilter, columnFilter,
\r
291 srcType, dstType, sumType, borderType ));
\r
295 void boxFilter( const Mat& src, Mat& dst, int ddepth,
\r
296 Size ksize, Point anchor,
\r
297 bool normalize, int borderType )
\r
299 int sdepth = src.depth(), cn = src.channels();
\r
302 dst.create( src.size(), CV_MAKETYPE(ddepth, cn) );
\r
303 if( borderType != BORDER_CONSTANT && normalize )
\r
305 if( src.rows == 1 )
\r
307 if( src.cols == 1 )
\r
310 Ptr<FilterEngine> f = createBoxFilter( src.type(), dst.type(),
\r
311 ksize, anchor, normalize, borderType );
\r
312 f->apply( src, dst );
\r
315 /****************************************************************************************\
\r
317 \****************************************************************************************/
\r
319 Mat getGaussianKernel( int n, double sigma, int ktype )
\r
321 const int SMALL_GAUSSIAN_SIZE = 7;
\r
322 static const float small_gaussian_tab[][SMALL_GAUSSIAN_SIZE] =
\r
325 {0.25f, 0.5f, 0.25f},
\r
326 {0.0625f, 0.25f, 0.375f, 0.25f, 0.0625f},
\r
327 {0.03125f, 0.109375f, 0.21875f, 0.28125f, 0.21875f, 0.109375f, 0.03125f}
\r
330 const float* fixed_kernel = n % 2 == 1 && n <= SMALL_GAUSSIAN_SIZE && sigma <= 0 ?
\r
331 small_gaussian_tab[n>>1] : 0;
\r
333 CV_Assert( ktype == CV_32F || ktype == CV_64F );
\r
334 Mat kernel(n, 1, ktype);
\r
335 float* cf = (float*)kernel.data;
\r
336 double* cd = (double*)kernel.data;
\r
338 double sigmaX = sigma > 0 ? sigma : ((n-1)*0.5 - 1)*0.3 + 0.8;
\r
339 double scale2X = -0.5/(sigmaX*sigmaX);
\r
343 for( i = 0; i < n; i++ )
\r
345 double x = i - (n-1)*0.5;
\r
346 double t = fixed_kernel ? (double)fixed_kernel[i] : std::exp(scale2X*x*x);
\r
347 if( ktype == CV_32F )
\r
360 for( i = 0; i < n; i++ )
\r
362 if( ktype == CV_32F )
\r
363 cf[i] = (float)(cf[i]*sum);
\r
372 Ptr<FilterEngine> createGaussianFilter( int type, Size ksize,
\r
373 double sigma1, double sigma2,
\r
376 int depth = CV_MAT_DEPTH(type);
\r
380 // automatic detection of kernel size from sigma
\r
381 if( ksize.width <= 0 && sigma1 > 0 )
\r
382 ksize.width = cvRound(sigma1*(depth == CV_8U ? 3 : 4)*2 + 1)|1;
\r
383 if( ksize.height <= 0 && sigma2 > 0 )
\r
384 ksize.height = cvRound(sigma2*(depth == CV_8U ? 3 : 4)*2 + 1)|1;
\r
386 CV_Assert( ksize.width > 0 && ksize.width % 2 == 1 &&
\r
387 ksize.height > 0 && ksize.height % 2 == 1 );
\r
389 sigma1 = std::max( sigma1, 0. );
\r
390 sigma2 = std::max( sigma2, 0. );
\r
392 Mat kx = getGaussianKernel( ksize.width, sigma1, std::max(depth, CV_32F) );
\r
394 if( ksize.height == ksize.width && std::abs(sigma1 - sigma2) < DBL_EPSILON )
\r
397 ky = getGaussianKernel( ksize.height, sigma2, std::max(depth, CV_32F) );
\r
399 return createSeparableLinearFilter( type, type, kx, ky, Point(-1,-1), 0, borderType );
\r
403 void GaussianBlur( const Mat& src, Mat& dst, Size ksize,
\r
404 double sigma1, double sigma2,
\r
407 if( ksize.width == 1 && ksize.height == 1 )
\r
413 dst.create( src.size(), src.type() );
\r
414 if( borderType != BORDER_CONSTANT )
\r
416 if( src.rows == 1 )
\r
418 if( src.cols == 1 )
\r
421 Ptr<FilterEngine> f = createGaussianFilter( src.type(), ksize, sigma1, sigma2, borderType );
\r
422 f->apply( src, dst );
\r
426 /****************************************************************************************\
\r
428 \****************************************************************************************/
\r
432 //#if defined __VEC__ || defined __ALTIVEC__
\r
433 //#define CV_ALTIVEC 1
\r
439 #include <altivec.h>
\r
443 #if _MSC_VER >= 1200
\r
444 #pragma warning( disable: 4244 )
\r
450 * This structure represents a two-tier histogram. The first tier (known as the
\r
451 * "coarse" level) is 4 bit wide and the second tier (known as the "fine" level)
\r
452 * is 8 bit wide. Pixels inserted in the fine level also get inserted into the
\r
453 * coarse bucket designated by the 4 MSBs of the fine bucket value.
\r
455 * The structure is aligned on 16 bits, which is a prerequisite for SIMD
\r
456 * instructions. Each bucket is 16 bit wide, which means that extra care must be
\r
457 * taken to prevent overflow.
\r
466 #if CV_SSE2 || defined __MMX__ || CV_ALTIVEC
\r
467 #define MEDIAN_HAVE_SIMD 1
\r
469 #define MEDIAN_HAVE_SIMD 0
\r
473 * histogram_add - adds histograms x and y.
\r
474 * histogram_sub - subtracts histogram x from y.
\r
477 static inline void histogram_add( const HT x[16], HT y[16] )
\r
479 const __m128i* rx = (const __m128i*)x;
\r
480 __m128i* ry = (__m128i*)y;
\r
481 __m128i r0 = _mm_add_epi16(_mm_load_si128(ry+0),_mm_load_si128(rx+0));
\r
482 __m128i r1 = _mm_add_epi16(_mm_load_si128(ry+1),_mm_load_si128(rx+1));
\r
483 _mm_store_si128(ry+0, r0);
\r
484 _mm_store_si128(ry+1, r1);
\r
487 static inline void histogram_sub( const HT x[16], HT y[16] )
\r
489 const __m128i* rx = (const __m128i*)x;
\r
490 __m128i* ry = (__m128i*)y;
\r
491 __m128i r0 = _mm_sub_epi16(_mm_load_si128(ry+0),_mm_load_si128(rx+0));
\r
492 __m128i r1 = _mm_sub_epi16(_mm_load_si128(ry+1),_mm_load_si128(rx+1));
\r
493 _mm_store_si128(ry+0, r0);
\r
494 _mm_store_si128(ry+1, r1);
\r
496 #elif defined(__MMX__)
\r
497 static inline void histogram_add( const HT x[16], HT y[16] )
\r
499 *(__m64*) &y[0] = _mm_add_pi16( *(__m64*) &y[0], *(__m64*) &x[0] );
\r
500 *(__m64*) &y[4] = _mm_add_pi16( *(__m64*) &y[4], *(__m64*) &x[4] );
\r
501 *(__m64*) &y[8] = _mm_add_pi16( *(__m64*) &y[8], *(__m64*) &x[8] );
\r
502 *(__m64*) &y[12] = _mm_add_pi16( *(__m64*) &y[12], *(__m64*) &x[12] );
\r
505 static inline void histogram_sub( const HT x[16], HT y[16] )
\r
507 *(__m64*) &y[0] = _mm_sub_pi16( *(__m64*) &y[0], *(__m64*) &x[0] );
\r
508 *(__m64*) &y[4] = _mm_sub_pi16( *(__m64*) &y[4], *(__m64*) &x[4] );
\r
509 *(__m64*) &y[8] = _mm_sub_pi16( *(__m64*) &y[8], *(__m64*) &x[8] );
\r
510 *(__m64*) &y[12] = _mm_sub_pi16( *(__m64*) &y[12], *(__m64*) &x[12] );
\r
513 static inline void histogram_add( const HT x[16], HT y[16] )
\r
515 *(vector HT*) &y[0] = vec_add( *(vector HT*) &y[0], *(vector HT*) &x[0] );
\r
516 *(vector HT*) &y[8] = vec_add( *(vector HT*) &y[8], *(vector HT*) &x[8] );
\r
519 static inline void histogram_sub( const HT x[16], HT y[16] )
\r
521 *(vector HT*) &y[0] = vec_sub( *(vector HT*) &y[0], *(vector HT*) &x[0] );
\r
522 *(vector HT*) &y[8] = vec_sub( *(vector HT*) &y[8], *(vector HT*) &x[8] );
\r
525 static inline void histogram_add( const HT x[16], HT y[16] )
\r
528 for( i = 0; i < 16; ++i )
\r
529 y[i] = (HT)(y[i] + x[i]);
\r
532 static inline void histogram_sub( const HT x[16], HT y[16] )
\r
535 for( i = 0; i < 16; ++i )
\r
536 y[i] = (HT)(y[i] - x[i]);
\r
540 static inline void histogram_muladd( int a, const HT x[16],
\r
543 for( int i = 0; i < 16; ++i )
\r
544 y[i] = (HT)(y[i] + a * x[i]);
\r
548 medianBlur_8u_O1( const Mat& _src, Mat& _dst, int ksize )
\r
551 * HOP is short for Histogram OPeration. This macro makes an operation \a op on
\r
552 * histogram \a h for pixel value \a x. It takes care of handling both levels.
\r
554 #define HOP(h,x,op) \
\r
555 h.coarse[x>>4] op, \
\r
556 *((HT*)h.fine + x) op
\r
558 #define COP(c,j,x,op) \
\r
559 h_coarse[ 16*(n*c+j) + (x>>4) ] op, \
\r
560 h_fine[ 16 * (n*(16*c+(x>>4)) + j) + (x & 0xF) ] op
\r
562 int cn = _dst.channels(), m = _dst.rows, r = (ksize-1)/2;
\r
563 size_t sstep = _src.step, dstep = _dst.step;
\r
564 Histogram CV_DECL_ALIGNED(16) H[4];
\r
567 int STRIPE_SIZE = std::min( _dst.cols, 512/cn );
\r
569 vector<HT> _h_coarse(1 * 16 * (STRIPE_SIZE + 2*r) * cn);
\r
570 vector<HT> _h_fine(16 * 16 * (STRIPE_SIZE + 2*r) * cn);
\r
571 HT* h_coarse = &_h_coarse[0];
\r
572 HT* h_fine = &_h_fine[0];
\r
574 for( int x = 0; x < _dst.cols; x += STRIPE_SIZE )
\r
576 int i, j, k, c, n = std::min(_dst.cols - x, STRIPE_SIZE) + r*2;
\r
577 const uchar* src = _src.data + x*cn;
\r
578 uchar* dst = _dst.data + (x - r)*cn;
\r
580 memset( h_coarse, 0, 16*n*cn*sizeof(h_coarse[0]) );
\r
581 memset( h_fine, 0, 16*16*n*cn*sizeof(h_fine[0]) );
\r
583 // First row initialization
\r
584 for( c = 0; c < cn; c++ )
\r
586 for( j = 0; j < n; j++ )
\r
587 COP( c, j, src[cn*j+c], += r+2 );
\r
589 for( i = 1; i < r; i++ )
\r
591 const uchar* p = src + sstep*std::min(i, m-1);
\r
592 for ( j = 0; j < n; j++ )
\r
593 COP( c, j, p[cn*j+c], ++ );
\r
597 for( i = 0; i < m; i++ )
\r
599 const uchar* p0 = src + sstep * std::max( 0, i-r-1 );
\r
600 const uchar* p1 = src + sstep * std::min( m-1, i+r );
\r
602 memset( H, 0, cn*sizeof(H[0]) );
\r
603 memset( luc, 0, cn*sizeof(luc[0]) );
\r
604 for( c = 0; c < cn; c++ )
\r
606 // Update column histograms for the entire row.
\r
607 for( j = 0; j < n; j++ )
\r
609 COP( c, j, p0[j*cn + c], -- );
\r
610 COP( c, j, p1[j*cn + c], ++ );
\r
613 // First column initialization
\r
614 for( j = 0; j < 2*r; ++j )
\r
615 histogram_add( &h_coarse[16*(n*c+j)], H[c].coarse );
\r
616 for( k = 0; k < 16; ++k )
\r
617 histogram_muladd( 2*r+1, &h_fine[16*n*(16*c+k)], &H[c].fine[k][0] );
\r
619 for( j = r; j < n-r; j++ )
\r
621 int t = 2*r*r + 2*r, b, sum = 0;
\r
624 histogram_add( &h_coarse[16*(n*c + std::min(j+r,n-1))], H[c].coarse );
\r
626 // Find median at coarse level
\r
627 for ( k = 0; k < 16 ; ++k )
\r
629 sum += H[c].coarse[k];
\r
632 sum -= H[c].coarse[k];
\r
638 /* Update corresponding histogram segment */
\r
639 if ( luc[c][k] <= j-r )
\r
641 memset( &H[c].fine[k], 0, 16 * sizeof(HT) );
\r
642 for ( luc[c][k] = j-r; luc[c][k] < MIN(j+r+1,n); ++luc[c][k] )
\r
643 histogram_add( &h_fine[16*(n*(16*c+k)+luc[c][k])], H[c].fine[k] );
\r
645 if ( luc[c][k] < j+r+1 )
\r
647 histogram_muladd( j+r+1 - n, &h_fine[16*(n*(16*c+k)+(n-1))], &H[c].fine[k][0] );
\r
648 luc[c][k] = (HT)(j+r+1);
\r
653 for ( ; luc[c][k] < j+r+1; ++luc[c][k] )
\r
655 histogram_sub( &h_fine[16*(n*(16*c+k)+MAX(luc[c][k]-2*r-1,0))], H[c].fine[k] );
\r
656 histogram_add( &h_fine[16*(n*(16*c+k)+MIN(luc[c][k],n-1))], H[c].fine[k] );
\r
660 histogram_sub( &h_coarse[16*(n*c+MAX(j-r,0))], H[c].coarse );
\r
662 /* Find median in segment */
\r
663 segment = H[c].fine[k];
\r
664 for ( b = 0; b < 16 ; b++ )
\r
669 dst[dstep*i+cn*j+c] = (uchar)(16*k + b);
\r
678 #if defined(__MMX__)
\r
687 #if _MSC_VER >= 1200
\r
688 #pragma warning( default: 4244 )
\r
692 medianBlur_8u_Om( const Mat& _src, Mat& _dst, int m )
\r
699 Size size = _dst.size();
\r
700 const uchar* src = _src.data;
\r
701 uchar* dst = _dst.data;
\r
702 int src_step = (int)_src.step, dst_step = (int)_dst.step;
\r
703 int cn = _src.channels();
\r
704 const uchar* src_max = src + size.height*src_step;
\r
706 #define UPDATE_ACC01( pix, cn, op ) \
\r
710 zone0[cn][p >> 4] op; \
\r
713 //CV_Assert( size.height >= nx && size.width >= nx );
\r
714 for( x = 0; x < size.width; x++, src += cn, dst += cn )
\r
716 uchar* dst_cur = dst;
\r
717 const uchar* src_top = src;
\r
718 const uchar* src_bottom = src;
\r
720 int src_step1 = src_step, dst_step1 = dst_step;
\r
724 src_bottom = src_top += src_step*(size.height-1);
\r
725 dst_cur += dst_step*(size.height-1);
\r
726 src_step1 = -src_step1;
\r
727 dst_step1 = -dst_step1;
\r
730 // init accumulator
\r
731 memset( zone0, 0, sizeof(zone0[0])*cn );
\r
732 memset( zone1, 0, sizeof(zone1[0])*cn );
\r
734 for( y = 0; y <= m/2; y++ )
\r
736 for( c = 0; c < cn; c++ )
\r
740 for( k = 0; k < m*cn; k += cn )
\r
741 UPDATE_ACC01( src_bottom[k+c], c, ++ );
\r
745 for( k = 0; k < m*cn; k += cn )
\r
746 UPDATE_ACC01( src_bottom[k+c], c, += m/2+1 );
\r
750 if( (src_step1 > 0 && y < size.height-1) ||
\r
751 (src_step1 < 0 && size.height-y-1 > 0) )
\r
752 src_bottom += src_step1;
\r
755 for( y = 0; y < size.height; y++, dst_cur += dst_step1 )
\r
758 for( c = 0; c < cn; c++ )
\r
761 for( k = 0; ; k++ )
\r
763 int t = s + zone0[c][k];
\r
764 if( t > n2 ) break;
\r
768 for( k *= N; ;k++ )
\r
771 if( s > n2 ) break;
\r
774 dst_cur[c] = (uchar)k;
\r
777 if( y+1 == size.height )
\r
782 for( k = 0; k < m; k++ )
\r
784 int p = src_top[k];
\r
785 int q = src_bottom[k];
\r
794 for( k = 0; k < m*3; k += 3 )
\r
796 UPDATE_ACC01( src_top[k], 0, -- );
\r
797 UPDATE_ACC01( src_top[k+1], 1, -- );
\r
798 UPDATE_ACC01( src_top[k+2], 2, -- );
\r
800 UPDATE_ACC01( src_bottom[k], 0, ++ );
\r
801 UPDATE_ACC01( src_bottom[k+1], 1, ++ );
\r
802 UPDATE_ACC01( src_bottom[k+2], 2, ++ );
\r
808 for( k = 0; k < m*4; k += 4 )
\r
810 UPDATE_ACC01( src_top[k], 0, -- );
\r
811 UPDATE_ACC01( src_top[k+1], 1, -- );
\r
812 UPDATE_ACC01( src_top[k+2], 2, -- );
\r
813 UPDATE_ACC01( src_top[k+3], 3, -- );
\r
815 UPDATE_ACC01( src_bottom[k], 0, ++ );
\r
816 UPDATE_ACC01( src_bottom[k+1], 1, ++ );
\r
817 UPDATE_ACC01( src_bottom[k+2], 2, ++ );
\r
818 UPDATE_ACC01( src_bottom[k+3], 3, ++ );
\r
822 if( (src_step1 > 0 && src_bottom + src_step1 < src_max) ||
\r
823 (src_step1 < 0 && src_bottom + src_step1 >= src) )
\r
824 src_bottom += src_step1;
\r
827 src_top += src_step1;
\r
837 typedef uchar value_type;
\r
838 typedef int arg_type;
\r
840 arg_type load(const uchar* ptr) { return *ptr; }
\r
841 void store(uchar* ptr, arg_type val) { *ptr = (uchar)val; }
\r
842 void operator()(arg_type& a, arg_type& b) const
\r
844 int t = CV_FAST_CAST_8U(a - b);
\r
851 typedef ushort value_type;
\r
852 typedef int arg_type;
\r
854 arg_type load(const ushort* ptr) { return *ptr; }
\r
855 void store(ushort* ptr, arg_type val) { *ptr = (ushort)val; }
\r
856 void operator()(arg_type& a, arg_type& b) const
\r
859 a = std::min(a, b);
\r
860 b = std::max(b, t);
\r
866 typedef float value_type;
\r
867 typedef float arg_type;
\r
869 arg_type load(const float* ptr) { return *ptr; }
\r
870 void store(float* ptr, arg_type val) { *ptr = val; }
\r
871 void operator()(arg_type& a, arg_type& b) const
\r
874 a = std::min(a, b);
\r
875 b = std::max(b, t);
\r
883 typedef uchar value_type;
\r
884 typedef __m128i arg_type;
\r
885 enum { SIZE = 16 };
\r
886 arg_type load(const uchar* ptr) { return _mm_loadu_si128((const __m128i*)ptr); }
\r
887 void store(uchar* ptr, arg_type val) { _mm_storeu_si128((__m128i*)ptr, val); }
\r
888 void operator()(arg_type& a, arg_type& b) const
\r
891 a = _mm_min_epu8(a, b);
\r
892 b = _mm_max_epu8(b, t);
\r
897 struct MinMaxVec16u
\r
899 typedef ushort value_type;
\r
900 typedef __m128i arg_type;
\r
902 arg_type load(const ushort* ptr) { return _mm_loadu_si128((const __m128i*)ptr); }
\r
903 void store(ushort* ptr, arg_type val) { _mm_storeu_si128((__m128i*)ptr, val); }
\r
904 void operator()(arg_type& a, arg_type& b) const
\r
906 arg_type t = _mm_subs_epu16(a, b);
\r
907 a = _mm_subs_epu16(a, t);
\r
908 b = _mm_adds_epu16(b, t);
\r
913 struct MinMaxVec32f
\r
915 typedef float value_type;
\r
916 typedef __m128 arg_type;
\r
918 arg_type load(const float* ptr) { return _mm_loadu_ps(ptr); }
\r
919 void store(float* ptr, arg_type val) { _mm_storeu_ps(ptr, val); }
\r
920 void operator()(arg_type& a, arg_type& b) const
\r
923 a = _mm_min_ps(a, b);
\r
924 b = _mm_max_ps(b, t);
\r
931 typedef MinMax8u MinMaxVec8u;
\r
932 typedef MinMax16u MinMaxVec16u;
\r
933 typedef MinMax32f MinMaxVec32f;
\r
937 template<class Op, class VecOp>
\r
939 medianBlur_SortNet( const Mat& _src, Mat& _dst, int m )
\r
941 typedef typename Op::value_type T;
\r
942 typedef typename Op::arg_type WT;
\r
943 typedef typename VecOp::arg_type VT;
\r
945 const T* src = (const T*)_src.data;
\r
946 T* dst = (T*)_dst.data;
\r
947 int sstep = (int)(_src.step/sizeof(T));
\r
948 int dstep = (int)(_dst.step/sizeof(T));
\r
949 Size size = _dst.size();
\r
950 int i, j, k, cn = _src.channels();
\r
956 if( size.width == 1 || size.height == 1 )
\r
958 int len = size.width + size.height - 1;
\r
959 int sdelta = size.height == 1 ? cn : sstep;
\r
960 int sdelta0 = size.height == 1 ? 0 : sstep - cn;
\r
961 int ddelta = size.height == 1 ? cn : dstep;
\r
963 for( i = 0; i < len; i++, src += sdelta0, dst += ddelta )
\r
964 for( j = 0; j < cn; j++, src++ )
\r
966 WT p0 = src[i > 0 ? -sdelta : 0];
\r
968 WT p2 = src[i < len - 1 ? sdelta : 0];
\r
970 op(p0, p1); op(p1, p2); op(p0, p1);
\r
977 for( i = 0; i < size.height; i++, dst += dstep )
\r
979 const T* row0 = src + std::max(i - 1, 0)*sstep;
\r
980 const T* row1 = src + i*sstep;
\r
981 const T* row2 = src + std::min(i + 1, size.height-1)*sstep;
\r
986 for( ; j < limit; j++ )
\r
988 int j0 = j >= cn ? j - cn : j;
\r
989 int j2 = j < size.width - cn ? j + cn : j;
\r
990 WT p0 = row0[j0], p1 = row0[j], p2 = row0[j2];
\r
991 WT p3 = row1[j0], p4 = row1[j], p5 = row1[j2];
\r
992 WT p6 = row2[j0], p7 = row2[j], p8 = row2[j2];
\r
994 op(p1, p2); op(p4, p5); op(p7, p8); op(p0, p1);
\r
995 op(p3, p4); op(p6, p7); op(p1, p2); op(p4, p5);
\r
996 op(p7, p8); op(p0, p3); op(p5, p8); op(p4, p7);
\r
997 op(p3, p6); op(p1, p4); op(p2, p5); op(p4, p7);
\r
998 op(p4, p2); op(p6, p4); op(p4, p2);
\r
1002 if( limit == size.width )
\r
1005 for( ; j <= size.width - VecOp::SIZE - cn; j += VecOp::SIZE )
\r
1007 VT p0 = vop.load(row0+j-cn), p1 = vop.load(row0+j), p2 = vop.load(row0+j+cn);
\r
1008 VT p3 = vop.load(row1+j-cn), p4 = vop.load(row1+j), p5 = vop.load(row1+j+cn);
\r
1009 VT p6 = vop.load(row2+j-cn), p7 = vop.load(row2+j), p8 = vop.load(row2+j+cn);
\r
1011 vop(p1, p2); vop(p4, p5); vop(p7, p8); vop(p0, p1);
\r
1012 vop(p3, p4); vop(p6, p7); vop(p1, p2); vop(p4, p5);
\r
1013 vop(p7, p8); vop(p0, p3); vop(p5, p8); vop(p4, p7);
\r
1014 vop(p3, p6); vop(p1, p4); vop(p2, p5); vop(p4, p7);
\r
1015 vop(p4, p2); vop(p6, p4); vop(p4, p2);
\r
1016 vop.store(dst+j, p4);
\r
1019 limit = size.width;
\r
1025 if( size.width == 1 || size.height == 1 )
\r
1027 int len = size.width + size.height - 1;
\r
1028 int sdelta = size.height == 1 ? cn : sstep;
\r
1029 int sdelta0 = size.height == 1 ? 0 : sstep - cn;
\r
1030 int ddelta = size.height == 1 ? cn : dstep;
\r
1032 for( i = 0; i < len; i++, src += sdelta0, dst += ddelta )
\r
1033 for( j = 0; j < cn; j++, src++ )
\r
1035 int i1 = i > 0 ? -sdelta : 0;
\r
1036 int i0 = i > 1 ? -sdelta*2 : i1;
\r
1037 int i3 = i < len-1 ? sdelta : 0;
\r
1038 int i4 = i < len-2 ? sdelta*2 : i3;
\r
1039 WT p0 = src[i0], p1 = src[i1], p2 = src[0], p3 = src[i3], p4 = src[i4];
\r
1041 op(p0, p1); op(p3, p4); op(p2, p3); op(p3, p4); op(p0, p2);
\r
1042 op(p2, p4); op(p1, p3); op(p1, p2);
\r
1049 for( i = 0; i < size.height; i++, dst += dstep )
\r
1052 row[0] = src + std::max(i - 2, 0)*sstep;
\r
1053 row[1] = src + std::max(i - 1, 0)*sstep;
\r
1054 row[2] = src + i*sstep;
\r
1055 row[3] = src + std::min(i + 1, size.height-1)*sstep;
\r
1056 row[4] = src + std::min(i + 2, size.height-1)*sstep;
\r
1061 for( ; j < limit; j++ )
\r
1064 int j1 = j >= cn ? j - cn : j;
\r
1065 int j0 = j >= cn*2 ? j - cn*2 : j1;
\r
1066 int j3 = j < size.width - cn ? j + cn : j;
\r
1067 int j4 = j < size.width - cn*2 ? j + cn*2 : j3;
\r
1068 for( k = 0; k < 5; k++ )
\r
1070 const T* rowk = row[k];
\r
1071 p[k*5] = rowk[j0]; p[k*5+1] = rowk[j1];
\r
1072 p[k*5+2] = rowk[j]; p[k*5+3] = rowk[j3];
\r
1073 p[k*5+4] = rowk[j4];
\r
1076 op(p[1], p[2]); op(p[0], p[1]); op(p[1], p[2]); op(p[4], p[5]); op(p[3], p[4]);
\r
1077 op(p[4], p[5]); op(p[0], p[3]); op(p[2], p[5]); op(p[2], p[3]); op(p[1], p[4]);
\r
1078 op(p[1], p[2]); op(p[3], p[4]); op(p[7], p[8]); op(p[6], p[7]); op(p[7], p[8]);
\r
1079 op(p[10], p[11]); op(p[9], p[10]); op(p[10], p[11]); op(p[6], p[9]); op(p[8], p[11]);
\r
1080 op(p[8], p[9]); op(p[7], p[10]); op(p[7], p[8]); op(p[9], p[10]); op(p[0], p[6]);
\r
1081 op(p[4], p[10]); op(p[4], p[6]); op(p[2], p[8]); op(p[2], p[4]); op(p[6], p[8]);
\r
1082 op(p[1], p[7]); op(p[5], p[11]); op(p[5], p[7]); op(p[3], p[9]); op(p[3], p[5]);
\r
1083 op(p[7], p[9]); op(p[1], p[2]); op(p[3], p[4]); op(p[5], p[6]); op(p[7], p[8]);
\r
1084 op(p[9], p[10]); op(p[13], p[14]); op(p[12], p[13]); op(p[13], p[14]); op(p[16], p[17]);
\r
1085 op(p[15], p[16]); op(p[16], p[17]); op(p[12], p[15]); op(p[14], p[17]); op(p[14], p[15]);
\r
1086 op(p[13], p[16]); op(p[13], p[14]); op(p[15], p[16]); op(p[19], p[20]); op(p[18], p[19]);
\r
1087 op(p[19], p[20]); op(p[21], p[22]); op(p[23], p[24]); op(p[21], p[23]); op(p[22], p[24]);
\r
1088 op(p[22], p[23]); op(p[18], p[21]); op(p[20], p[23]); op(p[20], p[21]); op(p[19], p[22]);
\r
1089 op(p[22], p[24]); op(p[19], p[20]); op(p[21], p[22]); op(p[23], p[24]); op(p[12], p[18]);
\r
1090 op(p[16], p[22]); op(p[16], p[18]); op(p[14], p[20]); op(p[20], p[24]); op(p[14], p[16]);
\r
1091 op(p[18], p[20]); op(p[22], p[24]); op(p[13], p[19]); op(p[17], p[23]); op(p[17], p[19]);
\r
1092 op(p[15], p[21]); op(p[15], p[17]); op(p[19], p[21]); op(p[13], p[14]); op(p[15], p[16]);
\r
1093 op(p[17], p[18]); op(p[19], p[20]); op(p[21], p[22]); op(p[23], p[24]); op(p[0], p[12]);
\r
1094 op(p[8], p[20]); op(p[8], p[12]); op(p[4], p[16]); op(p[16], p[24]); op(p[12], p[16]);
\r
1095 op(p[2], p[14]); op(p[10], p[22]); op(p[10], p[14]); op(p[6], p[18]); op(p[6], p[10]);
\r
1096 op(p[10], p[12]); op(p[1], p[13]); op(p[9], p[21]); op(p[9], p[13]); op(p[5], p[17]);
\r
1097 op(p[13], p[17]); op(p[3], p[15]); op(p[11], p[23]); op(p[11], p[15]); op(p[7], p[19]);
\r
1098 op(p[7], p[11]); op(p[11], p[13]); op(p[11], p[12]);
\r
1099 dst[j] = (T)p[12];
\r
1102 if( limit == size.width )
\r
1105 for( ; j <= size.width - VecOp::SIZE - cn*2; j += VecOp::SIZE )
\r
1108 for( k = 0; k < 5; k++ )
\r
1110 const T* rowk = row[k];
\r
1111 p[k*5] = vop.load(rowk+j-cn*2); p[k*5+1] = vop.load(rowk+j-cn);
\r
1112 p[k*5+2] = vop.load(rowk+j); p[k*5+3] = vop.load(rowk+j+cn);
\r
1113 p[k*5+4] = vop.load(rowk+j+cn*2);
\r
1116 vop(p[1], p[2]); vop(p[0], p[1]); vop(p[1], p[2]); vop(p[4], p[5]); vop(p[3], p[4]);
\r
1117 vop(p[4], p[5]); vop(p[0], p[3]); vop(p[2], p[5]); vop(p[2], p[3]); vop(p[1], p[4]);
\r
1118 vop(p[1], p[2]); vop(p[3], p[4]); vop(p[7], p[8]); vop(p[6], p[7]); vop(p[7], p[8]);
\r
1119 vop(p[10], p[11]); vop(p[9], p[10]); vop(p[10], p[11]); vop(p[6], p[9]); vop(p[8], p[11]);
\r
1120 vop(p[8], p[9]); vop(p[7], p[10]); vop(p[7], p[8]); vop(p[9], p[10]); vop(p[0], p[6]);
\r
1121 vop(p[4], p[10]); vop(p[4], p[6]); vop(p[2], p[8]); vop(p[2], p[4]); vop(p[6], p[8]);
\r
1122 vop(p[1], p[7]); vop(p[5], p[11]); vop(p[5], p[7]); vop(p[3], p[9]); vop(p[3], p[5]);
\r
1123 vop(p[7], p[9]); vop(p[1], p[2]); vop(p[3], p[4]); vop(p[5], p[6]); vop(p[7], p[8]);
\r
1124 vop(p[9], p[10]); vop(p[13], p[14]); vop(p[12], p[13]); vop(p[13], p[14]); vop(p[16], p[17]);
\r
1125 vop(p[15], p[16]); vop(p[16], p[17]); vop(p[12], p[15]); vop(p[14], p[17]); vop(p[14], p[15]);
\r
1126 vop(p[13], p[16]); vop(p[13], p[14]); vop(p[15], p[16]); vop(p[19], p[20]); vop(p[18], p[19]);
\r
1127 vop(p[19], p[20]); vop(p[21], p[22]); vop(p[23], p[24]); vop(p[21], p[23]); vop(p[22], p[24]);
\r
1128 vop(p[22], p[23]); vop(p[18], p[21]); vop(p[20], p[23]); vop(p[20], p[21]); vop(p[19], p[22]);
\r
1129 vop(p[22], p[24]); vop(p[19], p[20]); vop(p[21], p[22]); vop(p[23], p[24]); vop(p[12], p[18]);
\r
1130 vop(p[16], p[22]); vop(p[16], p[18]); vop(p[14], p[20]); vop(p[20], p[24]); vop(p[14], p[16]);
\r
1131 vop(p[18], p[20]); vop(p[22], p[24]); vop(p[13], p[19]); vop(p[17], p[23]); vop(p[17], p[19]);
\r
1132 vop(p[15], p[21]); vop(p[15], p[17]); vop(p[19], p[21]); vop(p[13], p[14]); vop(p[15], p[16]);
\r
1133 vop(p[17], p[18]); vop(p[19], p[20]); vop(p[21], p[22]); vop(p[23], p[24]); vop(p[0], p[12]);
\r
1134 vop(p[8], p[20]); vop(p[8], p[12]); vop(p[4], p[16]); vop(p[16], p[24]); vop(p[12], p[16]);
\r
1135 vop(p[2], p[14]); vop(p[10], p[22]); vop(p[10], p[14]); vop(p[6], p[18]); vop(p[6], p[10]);
\r
1136 vop(p[10], p[12]); vop(p[1], p[13]); vop(p[9], p[21]); vop(p[9], p[13]); vop(p[5], p[17]);
\r
1137 vop(p[13], p[17]); vop(p[3], p[15]); vop(p[11], p[23]); vop(p[11], p[15]); vop(p[7], p[19]);
\r
1138 vop(p[7], p[11]); vop(p[11], p[13]); vop(p[11], p[12]);
\r
1139 vop.store(dst+j, p[12]);
\r
1142 limit = size.width;
\r
1149 void medianBlur( const Mat& src0, Mat& dst, int ksize )
\r
1157 CV_Assert( ksize % 2 == 1 );
\r
1159 Size size = src0.size();
\r
1160 int cn = src0.channels();
\r
1161 bool useSortNet = ksize == 3 || (ksize == 5
\r
1163 && src0.depth() > CV_8U
\r
1167 dst.create( src0.size(), src0.type() );
\r
1171 if( dst.data != src0.data )
\r
1177 cv::copyMakeBorder( src0, src, 0, 0, ksize/2, ksize/2, BORDER_REPLICATE );
\r
1181 if( src.depth() == CV_8U )
\r
1182 medianBlur_SortNet<MinMax8u, MinMaxVec8u>( src, dst, ksize );
\r
1183 else if( src.depth() == CV_16U )
\r
1184 medianBlur_SortNet<MinMax16u, MinMaxVec16u>( src, dst, ksize );
\r
1185 else if( src.depth() == CV_32F )
\r
1186 medianBlur_SortNet<MinMax32f, MinMaxVec32f>( src, dst, ksize );
\r
1190 CV_Assert( src.depth() == CV_8U && (cn == 1 || cn == 3 || cn == 4) );
\r
1192 double img_size_mp = (double)(size.width*size.height)/(1 << 20);
\r
1193 if( ksize <= 3 + (img_size_mp < 1 ? 12 : img_size_mp < 4 ? 6 : 2)*(MEDIAN_HAVE_SIMD ? 1 : 3))
\r
1194 medianBlur_8u_Om( src, dst, ksize );
\r
1196 medianBlur_8u_O1( src, dst, ksize );
\r
1199 /****************************************************************************************\
\r
1200 Bilateral Filtering
\r
1201 \****************************************************************************************/
\r
1204 bilateralFilter_8u( const Mat& src, Mat& dst, int d,
\r
1205 double sigma_color, double sigma_space,
\r
1208 double gauss_color_coeff = -0.5/(sigma_color*sigma_color);
\r
1209 double gauss_space_coeff = -0.5/(sigma_space*sigma_space);
\r
1210 int cn = src.channels();
\r
1211 int i, j, k, maxk, radius;
\r
1212 Size size = src.size();
\r
1214 CV_Assert( (src.type() == CV_8UC1 || src.type() == CV_8UC3) &&
\r
1215 src.type() == dst.type() && src.size() == dst.size() &&
\r
1216 src.data != dst.data );
\r
1218 if( sigma_color <= 0 )
\r
1220 if( sigma_space <= 0 )
\r
1224 radius = cvRound(sigma_space*1.5);
\r
1227 radius = MAX(radius, 1);
\r
1231 copyMakeBorder( src, temp, radius, radius, radius, radius, borderType );
\r
1233 vector<float> _color_weight(cn*256);
\r
1234 vector<float> _space_weight(d*d);
\r
1235 vector<int> _space_ofs(d*d);
\r
1236 float* color_weight = &_color_weight[0];
\r
1237 float* space_weight = &_space_weight[0];
\r
1238 int* space_ofs = &_space_ofs[0];
\r
1240 // initialize color-related bilateral filter coefficients
\r
1241 for( i = 0; i < 256*cn; i++ )
\r
1242 color_weight[i] = (float)std::exp(i*i*gauss_color_coeff);
\r
1244 // initialize space-related bilateral filter coefficients
\r
1245 for( i = -radius, maxk = 0; i <= radius; i++ )
\r
1246 for( j = -radius; j <= radius; j++ )
\r
1248 double r = std::sqrt((double)i*i + (double)j*j);
\r
1251 space_weight[maxk] = (float)std::exp(r*r*gauss_space_coeff);
\r
1252 space_ofs[maxk++] = (int)(i*temp.step + j*cn);
\r
1255 for( i = 0; i < size.height; i++ )
\r
1257 const uchar* sptr = temp.data + (i+radius)*temp.step + radius*cn;
\r
1258 uchar* dptr = dst.data + i*dst.step;
\r
1262 for( j = 0; j < size.width; j++ )
\r
1264 float sum = 0, wsum = 0;
\r
1265 int val0 = sptr[j];
\r
1266 for( k = 0; k < maxk; k++ )
\r
1268 int val = sptr[j + space_ofs[k]];
\r
1269 float w = space_weight[k]*color_weight[std::abs(val - val0)];
\r
1273 // overflow is not possible here => there is no need to use CV_CAST_8U
\r
1274 dptr[j] = (uchar)cvRound(sum/wsum);
\r
1279 assert( cn == 3 );
\r
1280 for( j = 0; j < size.width*3; j += 3 )
\r
1282 float sum_b = 0, sum_g = 0, sum_r = 0, wsum = 0;
\r
1283 int b0 = sptr[j], g0 = sptr[j+1], r0 = sptr[j+2];
\r
1284 for( k = 0; k < maxk; k++ )
\r
1286 const uchar* sptr_k = sptr + j + space_ofs[k];
\r
1287 int b = sptr_k[0], g = sptr_k[1], r = sptr_k[2];
\r
1288 float w = space_weight[k]*color_weight[std::abs(b - b0) +
\r
1289 std::abs(g - g0) + std::abs(r - r0)];
\r
1290 sum_b += b*w; sum_g += g*w; sum_r += r*w;
\r
1294 b0 = cvRound(sum_b*wsum);
\r
1295 g0 = cvRound(sum_g*wsum);
\r
1296 r0 = cvRound(sum_r*wsum);
\r
1297 dptr[j] = (uchar)b0; dptr[j+1] = (uchar)g0; dptr[j+2] = (uchar)r0;
\r
1305 bilateralFilter_32f( const Mat& src, Mat& dst, int d,
\r
1306 double sigma_color, double sigma_space,
\r
1309 double gauss_color_coeff = -0.5/(sigma_color*sigma_color);
\r
1310 double gauss_space_coeff = -0.5/(sigma_space*sigma_space);
\r
1311 int cn = src.channels();
\r
1312 int i, j, k, maxk, radius;
\r
1313 double minValSrc=-1, maxValSrc=1;
\r
1314 const int kExpNumBinsPerChannel = 1 << 12;
\r
1315 int kExpNumBins = 0;
\r
1316 float lastExpVal = 1.f;
\r
1317 float len, scale_index;
\r
1318 Size size = src.size();
\r
1320 CV_Assert( (src.type() == CV_32FC1 || src.type() == CV_32FC3) &&
\r
1321 src.type() == dst.type() && src.size() == dst.size() &&
\r
1322 src.data != dst.data );
\r
1324 if( sigma_color <= 0 )
\r
1326 if( sigma_space <= 0 )
\r
1330 radius = cvRound(sigma_space*1.5);
\r
1333 radius = MAX(radius, 1);
\r
1335 // compute the min/max range for the input image (even if multichannel)
\r
1337 minMaxLoc( src.reshape(1), &minValSrc, &maxValSrc );
\r
1339 // temporary copy of the image with borders for easy processing
\r
1341 copyMakeBorder( src, temp, radius, radius, radius, radius, borderType );
\r
1343 // allocate lookup tables
\r
1344 vector<float> _space_weight(d*d);
\r
1345 vector<int> _space_ofs(d*d);
\r
1346 float* space_weight = &_space_weight[0];
\r
1347 int* space_ofs = &_space_ofs[0];
\r
1349 // assign a length which is slightly more than needed
\r
1350 len = (float)(maxValSrc - minValSrc) * cn;
\r
1351 kExpNumBins = kExpNumBinsPerChannel * cn;
\r
1352 vector<float> _expLUT(kExpNumBins+2);
\r
1353 float* expLUT = &_expLUT[0];
\r
1355 scale_index = kExpNumBins/len;
\r
1357 // initialize the exp LUT
\r
1358 for( i = 0; i < kExpNumBins+2; i++ )
\r
1360 if( lastExpVal > 0.f )
\r
1362 double val = i / scale_index;
\r
1363 expLUT[i] = (float)std::exp(val * val * gauss_color_coeff);
\r
1364 lastExpVal = expLUT[i];
\r
1370 // initialize space-related bilateral filter coefficients
\r
1371 for( i = -radius, maxk = 0; i <= radius; i++ )
\r
1372 for( j = -radius; j <= radius; j++ )
\r
1374 double r = std::sqrt((double)i*i + (double)j*j);
\r
1377 space_weight[maxk] = (float)std::exp(r*r*gauss_space_coeff);
\r
1378 space_ofs[maxk++] = (int)(i*(temp.step/sizeof(float)) + j*cn);
\r
1381 for( i = 0; i < size.height; i++ )
\r
1383 const float* sptr = (const float*)(temp.data + (i+radius)*temp.step) + radius*cn;
\r
1384 float* dptr = (float*)(dst.data + i*dst.step);
\r
1388 for( j = 0; j < size.width; j++ )
\r
1390 float sum = 0, wsum = 0;
\r
1391 float val0 = sptr[j];
\r
1392 for( k = 0; k < maxk; k++ )
\r
1394 float val = sptr[j + space_ofs[k]];
\r
1395 float alpha = (float)(std::abs(val - val0)*scale_index);
\r
1396 int idx = cvFloor(alpha);
\r
1398 float w = space_weight[k]*(expLUT[idx] + alpha*(expLUT[idx+1] - expLUT[idx]));
\r
1402 dptr[j] = (float)(sum/wsum);
\r
1407 assert( cn == 3 );
\r
1408 for( j = 0; j < size.width*3; j += 3 )
\r
1410 float sum_b = 0, sum_g = 0, sum_r = 0, wsum = 0;
\r
1411 float b0 = sptr[j], g0 = sptr[j+1], r0 = sptr[j+2];
\r
1412 for( k = 0; k < maxk; k++ )
\r
1414 const float* sptr_k = sptr + j + space_ofs[k];
\r
1415 float b = sptr_k[0], g = sptr_k[1], r = sptr_k[2];
\r
1416 float alpha = (float)((std::abs(b - b0) +
\r
1417 std::abs(g - g0) + std::abs(r - r0))*scale_index);
\r
1418 int idx = cvFloor(alpha);
\r
1420 float w = space_weight[k]*(expLUT[idx] + alpha*(expLUT[idx+1] - expLUT[idx]));
\r
1421 sum_b += b*w; sum_g += g*w; sum_r += r*w;
\r
1428 dptr[j] = b0; dptr[j+1] = g0; dptr[j+2] = r0;
\r
1435 void bilateralFilter( const Mat& src, Mat& dst, int d,
\r
1436 double sigmaColor, double sigmaSpace,
\r
1439 dst.create( src.size(), src.type() );
\r
1440 if( src.depth() == CV_8U )
\r
1441 bilateralFilter_8u( src, dst, d, sigmaColor, sigmaSpace, borderType );
\r
1442 else if( src.depth() == CV_32F )
\r
1443 bilateralFilter_32f( src, dst, d, sigmaColor, sigmaSpace, borderType );
\r
1445 CV_Error( CV_StsUnsupportedFormat,
\r
1446 "Bilateral filtering is only implemented for 8u and 32f images" );
\r
1451 //////////////////////////////////////////////////////////////////////////////////////////
\r
1454 cvSmooth( const void* srcarr, void* dstarr, int smooth_type,
\r
1455 int param1, int param2, double param3, double param4 )
\r
1457 cv::Mat src = cv::cvarrToMat(srcarr), dst0 = cv::cvarrToMat(dstarr), dst = dst0;
\r
1459 CV_Assert( dst.size() == src.size() &&
\r
1460 (smooth_type == CV_BLUR_NO_SCALE || dst.type() == src.type()) );
\r
1465 if( smooth_type == CV_BLUR || smooth_type == CV_BLUR_NO_SCALE )
\r
1466 cv::boxFilter( src, dst, dst.depth(), cv::Size(param1, param2), cv::Point(-1,-1),
\r
1467 smooth_type == CV_BLUR, cv::BORDER_REPLICATE );
\r
1468 else if( smooth_type == CV_GAUSSIAN )
\r
1469 cv::GaussianBlur( src, dst, cv::Size(param1, param2), param3, param4, cv::BORDER_REPLICATE );
\r
1470 else if( smooth_type == CV_MEDIAN )
\r
1471 cv::medianBlur( src, dst, param1 );
\r
1473 cv::bilateralFilter( src, dst, param1, param3, param4, cv::BORDER_REPLICATE );
\r
1475 if( dst.data != dst0.data )
\r
1476 CV_Error( CV_StsUnmatchedFormats, "The destination image does not have the proper type" );
\r
1479 /* End of file. */
\r