Home | History | Annotate | Download | only in src
      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 #include "precomp.hpp"
     42 #include "opencl_kernels_imgproc.hpp"
     43 
     44 namespace cv
     45 {
     46 
     47 // The function calculates center of gravity and the central second order moments
     48 static void completeMomentState( Moments* moments )
     49 {
     50     double cx = 0, cy = 0;
     51     double mu20, mu11, mu02;
     52     double inv_m00 = 0.0;
     53     assert( moments != 0 );
     54 
     55     if( fabs(moments->m00) > DBL_EPSILON )
     56     {
     57         inv_m00 = 1. / moments->m00;
     58         cx = moments->m10 * inv_m00;
     59         cy = moments->m01 * inv_m00;
     60     }
     61 
     62     // mu20 = m20 - m10*cx
     63     mu20 = moments->m20 - moments->m10 * cx;
     64     // mu11 = m11 - m10*cy
     65     mu11 = moments->m11 - moments->m10 * cy;
     66     // mu02 = m02 - m01*cy
     67     mu02 = moments->m02 - moments->m01 * cy;
     68 
     69     moments->mu20 = mu20;
     70     moments->mu11 = mu11;
     71     moments->mu02 = mu02;
     72 
     73     // mu30 = m30 - cx*(3*mu20 + cx*m10)
     74     moments->mu30 = moments->m30 - cx * (3 * mu20 + cx * moments->m10);
     75     mu11 += mu11;
     76     // mu21 = m21 - cx*(2*mu11 + cx*m01) - cy*mu20
     77     moments->mu21 = moments->m21 - cx * (mu11 + cx * moments->m01) - cy * mu20;
     78     // mu12 = m12 - cy*(2*mu11 + cy*m10) - cx*mu02
     79     moments->mu12 = moments->m12 - cy * (mu11 + cy * moments->m10) - cx * mu02;
     80     // mu03 = m03 - cy*(3*mu02 + cy*m01)
     81     moments->mu03 = moments->m03 - cy * (3 * mu02 + cy * moments->m01);
     82 
     83 
     84     double inv_sqrt_m00 = std::sqrt(std::abs(inv_m00));
     85     double s2 = inv_m00*inv_m00, s3 = s2*inv_sqrt_m00;
     86 
     87     moments->nu20 = moments->mu20*s2; moments->nu11 = moments->mu11*s2; moments->nu02 = moments->mu02*s2;
     88     moments->nu30 = moments->mu30*s3; moments->nu21 = moments->mu21*s3; moments->nu12 = moments->mu12*s3; moments->nu03 = moments->mu03*s3;
     89 
     90 }
     91 
     92 
     93 static Moments contourMoments( const Mat& contour )
     94 {
     95     Moments m;
     96     int lpt = contour.checkVector(2);
     97     int is_float = contour.depth() == CV_32F;
     98     const Point* ptsi = contour.ptr<Point>();
     99     const Point2f* ptsf = contour.ptr<Point2f>();
    100 
    101     CV_Assert( contour.depth() == CV_32S || contour.depth() == CV_32F );
    102 
    103     if( lpt == 0 )
    104         return m;
    105 
    106     double a00 = 0, a10 = 0, a01 = 0, a20 = 0, a11 = 0, a02 = 0, a30 = 0, a21 = 0, a12 = 0, a03 = 0;
    107     double xi, yi, xi2, yi2, xi_1, yi_1, xi_12, yi_12, dxy, xii_1, yii_1;
    108 
    109     if( !is_float )
    110     {
    111         xi_1 = ptsi[lpt-1].x;
    112         yi_1 = ptsi[lpt-1].y;
    113     }
    114     else
    115     {
    116         xi_1 = ptsf[lpt-1].x;
    117         yi_1 = ptsf[lpt-1].y;
    118     }
    119 
    120     xi_12 = xi_1 * xi_1;
    121     yi_12 = yi_1 * yi_1;
    122 
    123     for( int i = 0; i < lpt; i++ )
    124     {
    125         if( !is_float )
    126         {
    127             xi = ptsi[i].x;
    128             yi = ptsi[i].y;
    129         }
    130         else
    131         {
    132             xi = ptsf[i].x;
    133             yi = ptsf[i].y;
    134         }
    135 
    136         xi2 = xi * xi;
    137         yi2 = yi * yi;
    138         dxy = xi_1 * yi - xi * yi_1;
    139         xii_1 = xi_1 + xi;
    140         yii_1 = yi_1 + yi;
    141 
    142         a00 += dxy;
    143         a10 += dxy * xii_1;
    144         a01 += dxy * yii_1;
    145         a20 += dxy * (xi_1 * xii_1 + xi2);
    146         a11 += dxy * (xi_1 * (yii_1 + yi_1) + xi * (yii_1 + yi));
    147         a02 += dxy * (yi_1 * yii_1 + yi2);
    148         a30 += dxy * xii_1 * (xi_12 + xi2);
    149         a03 += dxy * yii_1 * (yi_12 + yi2);
    150         a21 += dxy * (xi_12 * (3 * yi_1 + yi) + 2 * xi * xi_1 * yii_1 +
    151                    xi2 * (yi_1 + 3 * yi));
    152         a12 += dxy * (yi_12 * (3 * xi_1 + xi) + 2 * yi * yi_1 * xii_1 +
    153                    yi2 * (xi_1 + 3 * xi));
    154         xi_1 = xi;
    155         yi_1 = yi;
    156         xi_12 = xi2;
    157         yi_12 = yi2;
    158     }
    159 
    160     if( fabs(a00) > FLT_EPSILON )
    161     {
    162         double db1_2, db1_6, db1_12, db1_24, db1_20, db1_60;
    163 
    164         if( a00 > 0 )
    165         {
    166             db1_2 = 0.5;
    167             db1_6 = 0.16666666666666666666666666666667;
    168             db1_12 = 0.083333333333333333333333333333333;
    169             db1_24 = 0.041666666666666666666666666666667;
    170             db1_20 = 0.05;
    171             db1_60 = 0.016666666666666666666666666666667;
    172         }
    173         else
    174         {
    175             db1_2 = -0.5;
    176             db1_6 = -0.16666666666666666666666666666667;
    177             db1_12 = -0.083333333333333333333333333333333;
    178             db1_24 = -0.041666666666666666666666666666667;
    179             db1_20 = -0.05;
    180             db1_60 = -0.016666666666666666666666666666667;
    181         }
    182 
    183         // spatial moments
    184         m.m00 = a00 * db1_2;
    185         m.m10 = a10 * db1_6;
    186         m.m01 = a01 * db1_6;
    187         m.m20 = a20 * db1_12;
    188         m.m11 = a11 * db1_24;
    189         m.m02 = a02 * db1_12;
    190         m.m30 = a30 * db1_20;
    191         m.m21 = a21 * db1_60;
    192         m.m12 = a12 * db1_60;
    193         m.m03 = a03 * db1_20;
    194 
    195         completeMomentState( &m );
    196     }
    197     return m;
    198 }
    199 
    200 
    201 /****************************************************************************************\
    202 *                                Spatial Raster Moments                                  *
    203 \****************************************************************************************/
    204 
    205 template<typename T, typename WT, typename MT>
    206 struct MomentsInTile_SIMD
    207 {
    208     int operator() (const T *, int, WT &, WT &, WT &, MT &)
    209     {
    210         return 0;
    211     }
    212 };
    213 
    214 #if CV_SSE2
    215 
    216 template <>
    217 struct MomentsInTile_SIMD<uchar, int, int>
    218 {
    219     MomentsInTile_SIMD()
    220     {
    221         useSIMD = checkHardwareSupport(CV_CPU_SSE2);
    222     }
    223 
    224     int operator() (const uchar * ptr, int len, int & x0, int & x1, int & x2, int & x3)
    225     {
    226         int x = 0;
    227 
    228         if( useSIMD )
    229         {
    230             __m128i qx_init = _mm_setr_epi16(0, 1, 2, 3, 4, 5, 6, 7);
    231             __m128i dx = _mm_set1_epi16(8);
    232             __m128i z = _mm_setzero_si128(), qx0 = z, qx1 = z, qx2 = z, qx3 = z, qx = qx_init;
    233 
    234             for( ; x <= len - 8; x += 8 )
    235             {
    236                 __m128i p = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i*)(ptr + x)), z);
    237                 __m128i sx = _mm_mullo_epi16(qx, qx);
    238 
    239                 qx0 = _mm_add_epi32(qx0, _mm_sad_epu8(p, z));
    240                 qx1 = _mm_add_epi32(qx1, _mm_madd_epi16(p, qx));
    241                 qx2 = _mm_add_epi32(qx2, _mm_madd_epi16(p, sx));
    242                 qx3 = _mm_add_epi32(qx3, _mm_madd_epi16( _mm_mullo_epi16(p, qx), sx));
    243 
    244                 qx = _mm_add_epi16(qx, dx);
    245             }
    246 
    247             _mm_store_si128((__m128i*)buf, qx0);
    248             x0 = buf[0] + buf[1] + buf[2] + buf[3];
    249             _mm_store_si128((__m128i*)buf, qx1);
    250             x1 = buf[0] + buf[1] + buf[2] + buf[3];
    251             _mm_store_si128((__m128i*)buf, qx2);
    252             x2 = buf[0] + buf[1] + buf[2] + buf[3];
    253             _mm_store_si128((__m128i*)buf, qx3);
    254             x3 = buf[0] + buf[1] + buf[2] + buf[3];
    255         }
    256 
    257         return x;
    258     }
    259 
    260     int CV_DECL_ALIGNED(16) buf[4];
    261     bool useSIMD;
    262 };
    263 
    264 #elif CV_NEON
    265 
    266 template <>
    267 struct MomentsInTile_SIMD<uchar, int, int>
    268 {
    269     MomentsInTile_SIMD()
    270     {
    271         ushort CV_DECL_ALIGNED(8) init[4] = { 0, 1, 2, 3 };
    272         qx_init = vld1_u16(init);
    273         v_step = vdup_n_u16(4);
    274     }
    275 
    276     int operator() (const uchar * ptr, int len, int & x0, int & x1, int & x2, int & x3)
    277     {
    278         int x = 0;
    279 
    280         uint32x4_t v_z = vdupq_n_u32(0), v_x0 = v_z, v_x1 = v_z,
    281             v_x2 = v_z, v_x3 = v_z;
    282         uint16x4_t qx = qx_init;
    283 
    284         for( ; x <= len - 8; x += 8 )
    285         {
    286             uint16x8_t v_src = vmovl_u8(vld1_u8(ptr + x));
    287 
    288             // first part
    289             uint32x4_t v_qx = vmovl_u16(qx);
    290             uint16x4_t v_p = vget_low_u16(v_src);
    291             uint32x4_t v_px = vmull_u16(qx, v_p);
    292 
    293             v_x0 = vaddw_u16(v_x0, v_p);
    294             v_x1 = vaddq_u32(v_x1, v_px);
    295             v_px = vmulq_u32(v_px, v_qx);
    296             v_x2 = vaddq_u32(v_x2, v_px);
    297             v_x3 = vaddq_u32(v_x3, vmulq_u32(v_px, v_qx));
    298             qx = vadd_u16(qx, v_step);
    299 
    300             // second part
    301             v_qx = vmovl_u16(qx);
    302             v_p = vget_high_u16(v_src);
    303             v_px = vmull_u16(qx, v_p);
    304 
    305             v_x0 = vaddw_u16(v_x0, v_p);
    306             v_x1 = vaddq_u32(v_x1, v_px);
    307             v_px = vmulq_u32(v_px, v_qx);
    308             v_x2 = vaddq_u32(v_x2, v_px);
    309             v_x3 = vaddq_u32(v_x3, vmulq_u32(v_px, v_qx));
    310 
    311             qx = vadd_u16(qx, v_step);
    312         }
    313 
    314         vst1q_u32(buf, v_x0);
    315         x0 = buf[0] + buf[1] + buf[2] + buf[3];
    316         vst1q_u32(buf, v_x1);
    317         x1 = buf[0] + buf[1] + buf[2] + buf[3];
    318         vst1q_u32(buf, v_x2);
    319         x2 = buf[0] + buf[1] + buf[2] + buf[3];
    320         vst1q_u32(buf, v_x3);
    321         x3 = buf[0] + buf[1] + buf[2] + buf[3];
    322 
    323         return x;
    324     }
    325 
    326     uint CV_DECL_ALIGNED(16) buf[4];
    327     uint16x4_t qx_init, v_step;
    328 };
    329 
    330 #endif
    331 
    332 #if CV_SSE4_1
    333 
    334 template <>
    335 struct MomentsInTile_SIMD<ushort, int, int64>
    336 {
    337     MomentsInTile_SIMD()
    338     {
    339         useSIMD = checkHardwareSupport(CV_CPU_SSE4_1);
    340     }
    341 
    342     int operator() (const ushort * ptr, int len, int & x0, int & x1, int & x2, int64 & x3)
    343     {
    344         int x = 0;
    345 
    346         if (useSIMD)
    347         {
    348             __m128i vx_init0 = _mm_setr_epi32(0, 1, 2, 3), vx_init1 = _mm_setr_epi32(4, 5, 6, 7),
    349                 v_delta = _mm_set1_epi32(8), v_zero = _mm_setzero_si128(), v_x0 = v_zero,
    350                 v_x1 = v_zero, v_x2 = v_zero, v_x3 = v_zero, v_ix0 = vx_init0, v_ix1 = vx_init1;
    351 
    352             for( ; x <= len - 8; x += 8 )
    353             {
    354                 __m128i v_src = _mm_loadu_si128((const __m128i *)(ptr + x));
    355                 __m128i v_src0 = _mm_unpacklo_epi16(v_src, v_zero), v_src1 = _mm_unpackhi_epi16(v_src, v_zero);
    356 
    357                 v_x0 = _mm_add_epi32(v_x0, _mm_add_epi32(v_src0, v_src1));
    358                 __m128i v_x1_0 = _mm_mullo_epi32(v_src0, v_ix0), v_x1_1 = _mm_mullo_epi32(v_src1, v_ix1);
    359                 v_x1 = _mm_add_epi32(v_x1, _mm_add_epi32(v_x1_0, v_x1_1));
    360 
    361                 __m128i v_2ix0 = _mm_mullo_epi32(v_ix0, v_ix0), v_2ix1 = _mm_mullo_epi32(v_ix1, v_ix1);
    362                 v_x2 = _mm_add_epi32(v_x2, _mm_add_epi32(_mm_mullo_epi32(v_2ix0, v_src0), _mm_mullo_epi32(v_2ix1, v_src1)));
    363 
    364                 __m128i t = _mm_add_epi32(_mm_mullo_epi32(v_2ix0, v_x1_0), _mm_mullo_epi32(v_2ix1, v_x1_1));
    365                 v_x3 = _mm_add_epi64(v_x3, _mm_add_epi64(_mm_unpacklo_epi32(t, v_zero), _mm_unpackhi_epi32(t, v_zero)));
    366 
    367                 v_ix0 = _mm_add_epi32(v_ix0, v_delta);
    368                 v_ix1 = _mm_add_epi32(v_ix1, v_delta);
    369             }
    370 
    371             _mm_store_si128((__m128i*)buf, v_x0);
    372             x0 = buf[0] + buf[1] + buf[2] + buf[3];
    373             _mm_store_si128((__m128i*)buf, v_x1);
    374             x1 = buf[0] + buf[1] + buf[2] + buf[3];
    375             _mm_store_si128((__m128i*)buf, v_x2);
    376             x2 = buf[0] + buf[1] + buf[2] + buf[3];
    377 
    378             _mm_store_si128((__m128i*)buf64, v_x3);
    379             x3 = buf64[0] + buf64[1];
    380         }
    381 
    382         return x;
    383     }
    384 
    385     int CV_DECL_ALIGNED(16) buf[4];
    386     int64 CV_DECL_ALIGNED(16) buf64[2];
    387     bool useSIMD;
    388 };
    389 
    390 #endif
    391 
    392 template<typename T, typename WT, typename MT>
    393 #if defined __GNUC__ && __GNUC__ == 4 && __GNUC_MINOR__ >= 5 && __GNUC_MINOR__ < 9
    394 // Workaround for http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60196
    395 __attribute__((optimize("no-tree-vectorize")))
    396 #endif
    397 static void momentsInTile( const Mat& img, double* moments )
    398 {
    399     Size size = img.size();
    400     int x, y;
    401     MT mom[10] = {0,0,0,0,0,0,0,0,0,0};
    402     MomentsInTile_SIMD<T, WT, MT> vop;
    403 
    404     for( y = 0; y < size.height; y++ )
    405     {
    406         const T* ptr = img.ptr<T>(y);
    407         WT x0 = 0, x1 = 0, x2 = 0;
    408         MT x3 = 0;
    409         x = vop(ptr, size.width, x0, x1, x2, x3);
    410 
    411         for( ; x < size.width; x++ )
    412         {
    413             WT p = ptr[x];
    414             WT xp = x * p, xxp;
    415 
    416             x0 += p;
    417             x1 += xp;
    418             xxp = xp * x;
    419             x2 += xxp;
    420             x3 += xxp * x;
    421         }
    422 
    423         WT py = y * x0, sy = y*y;
    424 
    425         mom[9] += ((MT)py) * sy;  // m03
    426         mom[8] += ((MT)x1) * sy;  // m12
    427         mom[7] += ((MT)x2) * y;  // m21
    428         mom[6] += x3;             // m30
    429         mom[5] += x0 * sy;        // m02
    430         mom[4] += x1 * y;         // m11
    431         mom[3] += x2;             // m20
    432         mom[2] += py;             // m01
    433         mom[1] += x1;             // m10
    434         mom[0] += x0;             // m00
    435     }
    436 
    437     for( x = 0; x < 10; x++ )
    438         moments[x] = (double)mom[x];
    439 }
    440 
    441 typedef void (*MomentsInTileFunc)(const Mat& img, double* moments);
    442 
    443 Moments::Moments()
    444 {
    445     m00 = m10 = m01 = m20 = m11 = m02 = m30 = m21 = m12 = m03 =
    446     mu20 = mu11 = mu02 = mu30 = mu21 = mu12 = mu03 =
    447     nu20 = nu11 = nu02 = nu30 = nu21 = nu12 = nu03 = 0.;
    448 }
    449 
    450 Moments::Moments( double _m00, double _m10, double _m01, double _m20, double _m11,
    451                   double _m02, double _m30, double _m21, double _m12, double _m03 )
    452 {
    453     m00 = _m00; m10 = _m10; m01 = _m01;
    454     m20 = _m20; m11 = _m11; m02 = _m02;
    455     m30 = _m30; m21 = _m21; m12 = _m12; m03 = _m03;
    456 
    457     double cx = 0, cy = 0, inv_m00 = 0;
    458     if( std::abs(m00) > DBL_EPSILON )
    459     {
    460         inv_m00 = 1./m00;
    461         cx = m10*inv_m00; cy = m01*inv_m00;
    462     }
    463 
    464     mu20 = m20 - m10*cx;
    465     mu11 = m11 - m10*cy;
    466     mu02 = m02 - m01*cy;
    467 
    468     mu30 = m30 - cx*(3*mu20 + cx*m10);
    469     mu21 = m21 - cx*(2*mu11 + cx*m01) - cy*mu20;
    470     mu12 = m12 - cy*(2*mu11 + cy*m10) - cx*mu02;
    471     mu03 = m03 - cy*(3*mu02 + cy*m01);
    472 
    473     double inv_sqrt_m00 = std::sqrt(std::abs(inv_m00));
    474     double s2 = inv_m00*inv_m00, s3 = s2*inv_sqrt_m00;
    475 
    476     nu20 = mu20*s2; nu11 = mu11*s2; nu02 = mu02*s2;
    477     nu30 = mu30*s3; nu21 = mu21*s3; nu12 = mu12*s3; nu03 = mu03*s3;
    478 }
    479 
    480 #ifdef HAVE_OPENCL
    481 
    482 static bool ocl_moments( InputArray _src, Moments& m, bool binary)
    483 {
    484     const int TILE_SIZE = 32;
    485     const int K = 10;
    486 
    487     ocl::Kernel k = ocl::Kernel("moments", ocl::imgproc::moments_oclsrc,
    488         format("-D TILE_SIZE=%d%s",
    489         TILE_SIZE,
    490         binary ? " -D OP_MOMENTS_BINARY" : ""));
    491 
    492     if( k.empty() )
    493         return false;
    494 
    495     UMat src = _src.getUMat();
    496     Size sz = src.size();
    497     int xtiles = (sz.width + TILE_SIZE-1)/TILE_SIZE;
    498     int ytiles = (sz.height + TILE_SIZE-1)/TILE_SIZE;
    499     int ntiles = xtiles*ytiles;
    500     UMat umbuf(1, ntiles*K, CV_32S);
    501 
    502     size_t globalsize[] = {xtiles, sz.height}, localsize[] = {1, TILE_SIZE};
    503     bool ok = k.args(ocl::KernelArg::ReadOnly(src),
    504                      ocl::KernelArg::PtrWriteOnly(umbuf),
    505                      xtiles).run(2, globalsize, localsize, true);
    506     if(!ok)
    507         return false;
    508     Mat mbuf = umbuf.getMat(ACCESS_READ);
    509     for( int i = 0; i < ntiles; i++ )
    510     {
    511         double x = (i % xtiles)*TILE_SIZE, y = (i / xtiles)*TILE_SIZE;
    512         const int* mom = mbuf.ptr<int>() + i*K;
    513         double xm = x * mom[0], ym = y * mom[0];
    514 
    515         // accumulate moments computed in each tile
    516 
    517         // + m00 ( = m00' )
    518         m.m00 += mom[0];
    519 
    520         // + m10 ( = m10' + x*m00' )
    521         m.m10 += mom[1] + xm;
    522 
    523         // + m01 ( = m01' + y*m00' )
    524         m.m01 += mom[2] + ym;
    525 
    526         // + m20 ( = m20' + 2*x*m10' + x*x*m00' )
    527         m.m20 += mom[3] + x * (mom[1] * 2 + xm);
    528 
    529         // + m11 ( = m11' + x*m01' + y*m10' + x*y*m00' )
    530         m.m11 += mom[4] + x * (mom[2] + ym) + y * mom[1];
    531 
    532         // + m02 ( = m02' + 2*y*m01' + y*y*m00' )
    533         m.m02 += mom[5] + y * (mom[2] * 2 + ym);
    534 
    535         // + m30 ( = m30' + 3*x*m20' + 3*x*x*m10' + x*x*x*m00' )
    536         m.m30 += mom[6] + x * (3. * mom[3] + x * (3. * mom[1] + xm));
    537 
    538         // + m21 ( = m21' + x*(2*m11' + 2*y*m10' + x*m01' + x*y*m00') + y*m20')
    539         m.m21 += mom[7] + x * (2 * (mom[4] + y * mom[1]) + x * (mom[2] + ym)) + y * mom[3];
    540 
    541         // + m12 ( = m12' + y*(2*m11' + 2*x*m01' + y*m10' + x*y*m00') + x*m02')
    542         m.m12 += mom[8] + y * (2 * (mom[4] + x * mom[2]) + y * (mom[1] + xm)) + x * mom[5];
    543 
    544         // + m03 ( = m03' + 3*y*m02' + 3*y*y*m01' + y*y*y*m00' )
    545         m.m03 += mom[9] + y * (3. * mom[5] + y * (3. * mom[2] + ym));
    546     }
    547 
    548     return true;
    549 }
    550 
    551 #endif
    552 
    553 }
    554 
    555 
    556 cv::Moments cv::moments( InputArray _src, bool binary )
    557 {
    558     const int TILE_SIZE = 32;
    559     MomentsInTileFunc func = 0;
    560     uchar nzbuf[TILE_SIZE*TILE_SIZE];
    561     Moments m;
    562     int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
    563     Size size = _src.size();
    564 
    565     if( size.width <= 0 || size.height <= 0 )
    566         return m;
    567 
    568 #ifdef HAVE_OPENCL
    569     if( !(ocl::useOpenCL() && type == CV_8UC1  &&
    570         _src.isUMat() && ocl_moments(_src, m, binary)) )
    571 #endif
    572     {
    573         Mat mat = _src.getMat();
    574         if( mat.checkVector(2) >= 0 && (depth == CV_32F || depth == CV_32S))
    575             return contourMoments(mat);
    576 
    577         if( cn > 1 )
    578             CV_Error( CV_StsBadArg, "Invalid image type (must be single-channel)" );
    579 
    580 #if IPP_VERSION_X100 >= 801 && 0
    581         CV_IPP_CHECK()
    582         {
    583             if (!binary)
    584             {
    585                 IppiSize roi = { mat.cols, mat.rows };
    586                 IppiMomentState_64f * moment = NULL;
    587                 // ippiMomentInitAlloc_64f, ippiMomentFree_64f are deprecated in 8.1, but there are not another way
    588                 // to initialize IppiMomentState_64f. When GetStateSize and Init functions will appear we have to
    589                 // change our code.
    590                 CV_SUPPRESS_DEPRECATED_START
    591                 if (ippiMomentInitAlloc_64f(&moment, ippAlgHintAccurate) >= 0)
    592                 {
    593                     typedef IppStatus (CV_STDCALL * ippiMoments)(const void * pSrc, int srcStep, IppiSize roiSize, IppiMomentState_64f* pCtx);
    594                     ippiMoments ippFunc =
    595                         type == CV_8UC1 ? (ippiMoments)ippiMoments64f_8u_C1R :
    596                         type == CV_16UC1 ? (ippiMoments)ippiMoments64f_16u_C1R :
    597                         type == CV_32FC1? (ippiMoments)ippiMoments64f_32f_C1R : 0;
    598 
    599                     if (ippFunc)
    600                     {
    601                         if (ippFunc(mat.data, (int)mat.step, roi, moment) >= 0)
    602                         {
    603                             IppiPoint point = { 0, 0 };
    604                             ippiGetSpatialMoment_64f(moment, 0, 0, 0, point, &m.m00);
    605                             ippiGetSpatialMoment_64f(moment, 1, 0, 0, point, &m.m10);
    606                             ippiGetSpatialMoment_64f(moment, 0, 1, 0, point, &m.m01);
    607 
    608                             ippiGetSpatialMoment_64f(moment, 2, 0, 0, point, &m.m20);
    609                             ippiGetSpatialMoment_64f(moment, 1, 1, 0, point, &m.m11);
    610                             ippiGetSpatialMoment_64f(moment, 0, 2, 0, point, &m.m02);
    611 
    612                             ippiGetSpatialMoment_64f(moment, 3, 0, 0, point, &m.m30);
    613                             ippiGetSpatialMoment_64f(moment, 2, 1, 0, point, &m.m21);
    614                             ippiGetSpatialMoment_64f(moment, 1, 2, 0, point, &m.m12);
    615                             ippiGetSpatialMoment_64f(moment, 0, 3, 0, point, &m.m03);
    616                             ippiGetCentralMoment_64f(moment, 2, 0, 0, &m.mu20);
    617                             ippiGetCentralMoment_64f(moment, 1, 1, 0, &m.mu11);
    618                             ippiGetCentralMoment_64f(moment, 0, 2, 0, &m.mu02);
    619                             ippiGetCentralMoment_64f(moment, 3, 0, 0, &m.mu30);
    620                             ippiGetCentralMoment_64f(moment, 2, 1, 0, &m.mu21);
    621                             ippiGetCentralMoment_64f(moment, 1, 2, 0, &m.mu12);
    622                             ippiGetCentralMoment_64f(moment, 0, 3, 0, &m.mu03);
    623                             ippiGetNormalizedCentralMoment_64f(moment, 2, 0, 0, &m.nu20);
    624                             ippiGetNormalizedCentralMoment_64f(moment, 1, 1, 0, &m.nu11);
    625                             ippiGetNormalizedCentralMoment_64f(moment, 0, 2, 0, &m.nu02);
    626                             ippiGetNormalizedCentralMoment_64f(moment, 3, 0, 0, &m.nu30);
    627                             ippiGetNormalizedCentralMoment_64f(moment, 2, 1, 0, &m.nu21);
    628                             ippiGetNormalizedCentralMoment_64f(moment, 1, 2, 0, &m.nu12);
    629                             ippiGetNormalizedCentralMoment_64f(moment, 0, 3, 0, &m.nu03);
    630 
    631                             ippiMomentFree_64f(moment);
    632                             CV_IMPL_ADD(CV_IMPL_IPP);
    633                             return m;
    634                         }
    635                         setIppErrorStatus();
    636                     }
    637                     ippiMomentFree_64f(moment);
    638                 }
    639                 else
    640                     setIppErrorStatus();
    641                 CV_SUPPRESS_DEPRECATED_END
    642             }
    643         }
    644 #endif
    645 
    646         if( binary || depth == CV_8U )
    647             func = momentsInTile<uchar, int, int>;
    648         else if( depth == CV_16U )
    649             func = momentsInTile<ushort, int, int64>;
    650         else if( depth == CV_16S )
    651             func = momentsInTile<short, int, int64>;
    652         else if( depth == CV_32F )
    653             func = momentsInTile<float, double, double>;
    654         else if( depth == CV_64F )
    655             func = momentsInTile<double, double, double>;
    656         else
    657             CV_Error( CV_StsUnsupportedFormat, "" );
    658 
    659         Mat src0(mat);
    660 
    661         for( int y = 0; y < size.height; y += TILE_SIZE )
    662         {
    663             Size tileSize;
    664             tileSize.height = std::min(TILE_SIZE, size.height - y);
    665 
    666             for( int x = 0; x < size.width; x += TILE_SIZE )
    667             {
    668                 tileSize.width = std::min(TILE_SIZE, size.width - x);
    669                 Mat src(src0, cv::Rect(x, y, tileSize.width, tileSize.height));
    670 
    671                 if( binary )
    672                 {
    673                     cv::Mat tmp(tileSize, CV_8U, nzbuf);
    674                     cv::compare( src, 0, tmp, CV_CMP_NE );
    675                     src = tmp;
    676                 }
    677 
    678                 double mom[10];
    679                 func( src, mom );
    680 
    681                 if(binary)
    682                 {
    683                     double s = 1./255;
    684                     for( int k = 0; k < 10; k++ )
    685                         mom[k] *= s;
    686                 }
    687 
    688                 double xm = x * mom[0], ym = y * mom[0];
    689 
    690                 // accumulate moments computed in each tile
    691 
    692                 // + m00 ( = m00' )
    693                 m.m00 += mom[0];
    694 
    695                 // + m10 ( = m10' + x*m00' )
    696                 m.m10 += mom[1] + xm;
    697 
    698                 // + m01 ( = m01' + y*m00' )
    699                 m.m01 += mom[2] + ym;
    700 
    701                 // + m20 ( = m20' + 2*x*m10' + x*x*m00' )
    702                 m.m20 += mom[3] + x * (mom[1] * 2 + xm);
    703 
    704                 // + m11 ( = m11' + x*m01' + y*m10' + x*y*m00' )
    705                 m.m11 += mom[4] + x * (mom[2] + ym) + y * mom[1];
    706 
    707                 // + m02 ( = m02' + 2*y*m01' + y*y*m00' )
    708                 m.m02 += mom[5] + y * (mom[2] * 2 + ym);
    709 
    710                 // + m30 ( = m30' + 3*x*m20' + 3*x*x*m10' + x*x*x*m00' )
    711                 m.m30 += mom[6] + x * (3. * mom[3] + x * (3. * mom[1] + xm));
    712 
    713                 // + m21 ( = m21' + x*(2*m11' + 2*y*m10' + x*m01' + x*y*m00') + y*m20')
    714                 m.m21 += mom[7] + x * (2 * (mom[4] + y * mom[1]) + x * (mom[2] + ym)) + y * mom[3];
    715 
    716                 // + m12 ( = m12' + y*(2*m11' + 2*x*m01' + y*m10' + x*y*m00') + x*m02')
    717                 m.m12 += mom[8] + y * (2 * (mom[4] + x * mom[2]) + y * (mom[1] + xm)) + x * mom[5];
    718 
    719                 // + m03 ( = m03' + 3*y*m02' + 3*y*y*m01' + y*y*y*m00' )
    720                 m.m03 += mom[9] + y * (3. * mom[5] + y * (3. * mom[2] + ym));
    721             }
    722         }
    723     }
    724 
    725     completeMomentState( &m );
    726     return m;
    727 }
    728 
    729 
    730 void cv::HuMoments( const Moments& m, double hu[7] )
    731 {
    732     double t0 = m.nu30 + m.nu12;
    733     double t1 = m.nu21 + m.nu03;
    734 
    735     double q0 = t0 * t0, q1 = t1 * t1;
    736 
    737     double n4 = 4 * m.nu11;
    738     double s = m.nu20 + m.nu02;
    739     double d = m.nu20 - m.nu02;
    740 
    741     hu[0] = s;
    742     hu[1] = d * d + n4 * m.nu11;
    743     hu[3] = q0 + q1;
    744     hu[5] = d * (q0 - q1) + n4 * t0 * t1;
    745 
    746     t0 *= q0 - 3 * q1;
    747     t1 *= 3 * q0 - q1;
    748 
    749     q0 = m.nu30 - 3 * m.nu12;
    750     q1 = 3 * m.nu21 - m.nu03;
    751 
    752     hu[2] = q0 * q0 + q1 * q1;
    753     hu[4] = q0 * t0 + q1 * t1;
    754     hu[6] = q1 * t0 - q0 * t1;
    755 }
    756 
    757 void cv::HuMoments( const Moments& m, OutputArray _hu )
    758 {
    759     _hu.create(7, 1, CV_64F);
    760     Mat hu = _hu.getMat();
    761     CV_Assert( hu.isContinuous() );
    762     HuMoments(m, hu.ptr<double>());
    763 }
    764 
    765 
    766 CV_IMPL void cvMoments( const CvArr* arr, CvMoments* moments, int binary )
    767 {
    768     const IplImage* img = (const IplImage*)arr;
    769     cv::Mat src;
    770     if( CV_IS_IMAGE(arr) && img->roi && img->roi->coi > 0 )
    771         cv::extractImageCOI(arr, src, img->roi->coi-1);
    772     else
    773         src = cv::cvarrToMat(arr);
    774     cv::Moments m = cv::moments(src, binary != 0);
    775     CV_Assert( moments != 0 );
    776     *moments = m;
    777 }
    778 
    779 
    780 CV_IMPL double cvGetSpatialMoment( CvMoments * moments, int x_order, int y_order )
    781 {
    782     int order = x_order + y_order;
    783 
    784     if( !moments )
    785         CV_Error( CV_StsNullPtr, "" );
    786     if( (x_order | y_order) < 0 || order > 3 )
    787         CV_Error( CV_StsOutOfRange, "" );
    788 
    789     return (&(moments->m00))[order + (order >> 1) + (order > 2) * 2 + y_order];
    790 }
    791 
    792 
    793 CV_IMPL double cvGetCentralMoment( CvMoments * moments, int x_order, int y_order )
    794 {
    795     int order = x_order + y_order;
    796 
    797     if( !moments )
    798         CV_Error( CV_StsNullPtr, "" );
    799     if( (x_order | y_order) < 0 || order > 3 )
    800         CV_Error( CV_StsOutOfRange, "" );
    801 
    802     return order >= 2 ? (&(moments->m00))[4 + order * 3 + y_order] :
    803     order == 0 ? moments->m00 : 0;
    804 }
    805 
    806 
    807 CV_IMPL double cvGetNormalizedCentralMoment( CvMoments * moments, int x_order, int y_order )
    808 {
    809     int order = x_order + y_order;
    810 
    811     double mu = cvGetCentralMoment( moments, x_order, y_order );
    812     double m00s = moments->inv_sqrt_m00;
    813 
    814     while( --order >= 0 )
    815         mu *= m00s;
    816     return mu * m00s * m00s;
    817 }
    818 
    819 
    820 CV_IMPL void cvGetHuMoments( CvMoments * mState, CvHuMoments * HuState )
    821 {
    822     if( !mState || !HuState )
    823         CV_Error( CV_StsNullPtr, "" );
    824 
    825     double m00s = mState->inv_sqrt_m00, m00 = m00s * m00s, s2 = m00 * m00, s3 = s2 * m00s;
    826 
    827     double nu20 = mState->mu20 * s2,
    828     nu11 = mState->mu11 * s2,
    829     nu02 = mState->mu02 * s2,
    830     nu30 = mState->mu30 * s3,
    831     nu21 = mState->mu21 * s3, nu12 = mState->mu12 * s3, nu03 = mState->mu03 * s3;
    832 
    833     double t0 = nu30 + nu12;
    834     double t1 = nu21 + nu03;
    835 
    836     double q0 = t0 * t0, q1 = t1 * t1;
    837 
    838     double n4 = 4 * nu11;
    839     double s = nu20 + nu02;
    840     double d = nu20 - nu02;
    841 
    842     HuState->hu1 = s;
    843     HuState->hu2 = d * d + n4 * nu11;
    844     HuState->hu4 = q0 + q1;
    845     HuState->hu6 = d * (q0 - q1) + n4 * t0 * t1;
    846 
    847     t0 *= q0 - 3 * q1;
    848     t1 *= 3 * q0 - q1;
    849 
    850     q0 = nu30 - 3 * nu12;
    851     q1 = 3 * nu21 - nu03;
    852 
    853     HuState->hu3 = q0 * q0 + q1 * q1;
    854     HuState->hu5 = q0 * t0 + q1 * t1;
    855     HuState->hu7 = q1 * t0 - q0 * t1;
    856 }
    857 
    858 
    859 /* End of file. */
    860