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 //                           License Agreement
     11 //                For Open Source Computer Vision Library
     12 //
     13 // Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
     14 // Copyright (C) 2009, Willow Garage Inc., all rights reserved.
     15 // Third party copyrights are property of their respective owners.
     16 //
     17 // Redistribution and use in source and binary forms, with or without modification,
     18 // are permitted provided that the following conditions are met:
     19 //
     20 //   * Redistribution's of source code must retain the above copyright notice,
     21 //     this list of conditions and the following disclaimer.
     22 //
     23 //   * Redistribution's in binary form must reproduce the above copyright notice,
     24 //     this list of conditions and the following disclaimer in the documentation
     25 //     and/or other materials provided with the distribution.
     26 //
     27 //   * The name of the copyright holders may not be used to endorse or promote products
     28 //     derived from this software without specific prior written permission.
     29 //
     30 // This software is provided by the copyright holders and contributors "as is" and
     31 // any express or implied warranties, including, but not limited to, the implied
     32 // warranties of merchantability and fitness for a particular purpose are disclaimed.
     33 // In no event shall the Intel Corporation or contributors be liable for any direct,
     34 // indirect, incidental, special, exemplary, or consequential damages
     35 // (including, but not limited to, procurement of substitute goods or services;
     36 // loss of use, data, or profits; or business interruption) however caused
     37 // and on any theory of liability, whether in contract, strict liability,
     38 // or tort (including negligence or otherwise) arising in any way out of
     39 // the use of this software, even if advised of the possibility of such damage.
     40 //
     41 //M*/
     42 #include "precomp.hpp"
     43 
     44 #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
     45 
     46 void cv::cuda::meanShiftSegmentation(InputArray, OutputArray, int, int, int, TermCriteria, Stream&) { throw_no_cuda(); }
     47 
     48 #else
     49 
     50 // Auxiliray stuff
     51 namespace
     52 {
     53 
     54 //
     55 // Declarations
     56 //
     57 
     58 class DjSets
     59 {
     60 public:
     61     DjSets(int n);
     62     int find(int elem);
     63     int merge(int set1, int set2);
     64 
     65     std::vector<int> parent;
     66     std::vector<int> rank;
     67     std::vector<int> size;
     68 private:
     69     DjSets(const DjSets&);
     70     void operator =(const DjSets&);
     71 };
     72 
     73 
     74 template <typename T>
     75 struct GraphEdge
     76 {
     77     GraphEdge() {}
     78     GraphEdge(int to_, int next_, const T& val_) : to(to_), next(next_), val(val_) {}
     79     int to;
     80     int next;
     81     T val;
     82 };
     83 
     84 
     85 template <typename T>
     86 class Graph
     87 {
     88 public:
     89     typedef GraphEdge<T> Edge;
     90 
     91     Graph(int numv, int nume_max);
     92 
     93     void addEdge(int from, int to, const T& val=T());
     94 
     95     std::vector<int> start;
     96     std::vector<Edge> edges;
     97 
     98     int numv;
     99     int nume_max;
    100     int nume;
    101 private:
    102     Graph(const Graph&);
    103     void operator =(const Graph&);
    104 };
    105 
    106 
    107 struct SegmLinkVal
    108 {
    109     SegmLinkVal() {}
    110     SegmLinkVal(int dr_, int dsp_) : dr(dr_), dsp(dsp_) {}
    111     bool operator <(const SegmLinkVal& other) const
    112     {
    113         return dr + dsp < other.dr + other.dsp;
    114     }
    115     int dr;
    116     int dsp;
    117 };
    118 
    119 
    120 struct SegmLink
    121 {
    122     SegmLink() {}
    123     SegmLink(int from_, int to_, const SegmLinkVal& val_)
    124         : from(from_), to(to_), val(val_) {}
    125     bool operator <(const SegmLink& other) const
    126     {
    127         return val < other.val;
    128     }
    129     int from;
    130     int to;
    131     SegmLinkVal val;
    132 };
    133 
    134 //
    135 // Implementation
    136 //
    137 
    138 DjSets::DjSets(int n) : parent(n), rank(n, 0), size(n, 1)
    139 {
    140     for (int i = 0; i < n; ++i)
    141         parent[i] = i;
    142 }
    143 
    144 
    145 inline int DjSets::find(int elem)
    146 {
    147     int set = elem;
    148     while (set != parent[set])
    149         set = parent[set];
    150     while (elem != parent[elem])
    151     {
    152         int next = parent[elem];
    153         parent[elem] = set;
    154         elem = next;
    155     }
    156     return set;
    157 }
    158 
    159 
    160 inline int DjSets::merge(int set1, int set2)
    161 {
    162     if (rank[set1] < rank[set2])
    163     {
    164         parent[set1] = set2;
    165         size[set2] += size[set1];
    166         return set2;
    167     }
    168     if (rank[set2] < rank[set1])
    169     {
    170         parent[set2] = set1;
    171         size[set1] += size[set2];
    172         return set1;
    173     }
    174     parent[set1] = set2;
    175     rank[set2]++;
    176     size[set2] += size[set1];
    177     return set2;
    178 }
    179 
    180 
    181 template <typename T>
    182 Graph<T>::Graph(int numv_, int nume_max_) : start(numv_, -1), edges(nume_max_)
    183 {
    184     this->numv = numv_;
    185     this->nume_max = nume_max_;
    186     nume = 0;
    187 }
    188 
    189 
    190 template <typename T>
    191 inline void Graph<T>::addEdge(int from, int to, const T& val)
    192 {
    193     edges[nume] = Edge(to, start[from], val);
    194     start[from] = nume;
    195     nume++;
    196 }
    197 
    198 
    199 inline int pix(int y, int x, int ncols)
    200 {
    201     return y * ncols + x;
    202 }
    203 
    204 
    205 inline int sqr(int x)
    206 {
    207     return x * x;
    208 }
    209 
    210 
    211 inline int dist2(const cv::Vec4b& lhs, const cv::Vec4b& rhs)
    212 {
    213     return sqr(lhs[0] - rhs[0]) + sqr(lhs[1] - rhs[1]) + sqr(lhs[2] - rhs[2]);
    214 }
    215 
    216 
    217 inline int dist2(const cv::Vec2s& lhs, const cv::Vec2s& rhs)
    218 {
    219     return sqr(lhs[0] - rhs[0]) + sqr(lhs[1] - rhs[1]);
    220 }
    221 
    222 } // anonymous namespace
    223 
    224 
    225 void cv::cuda::meanShiftSegmentation(InputArray _src, OutputArray _dst, int sp, int sr, int minsize, TermCriteria criteria, Stream& stream)
    226 {
    227     GpuMat src = _src.getGpuMat();
    228 
    229     CV_Assert( src.type() == CV_8UC4 );
    230 
    231     const int nrows = src.rows;
    232     const int ncols = src.cols;
    233     const int hr = sr;
    234     const int hsp = sp;
    235 
    236     // Perform mean shift procedure and obtain region and spatial maps
    237     GpuMat d_rmap, d_spmap;
    238     cuda::meanShiftProc(src, d_rmap, d_spmap, sp, sr, criteria, stream);
    239 
    240     stream.waitForCompletion();
    241 
    242     Mat rmap(d_rmap);
    243     Mat spmap(d_spmap);
    244 
    245     Graph<SegmLinkVal> g(nrows * ncols, 4 * (nrows - 1) * (ncols - 1)
    246                                         + (nrows - 1) + (ncols - 1));
    247 
    248     // Make region adjacent graph from image
    249     Vec4b r1;
    250     Vec4b r2[4];
    251     Vec2s sp1;
    252     Vec2s sp2[4];
    253     int dr[4];
    254     int dsp[4];
    255     for (int y = 0; y < nrows - 1; ++y)
    256     {
    257         Vec4b* ry = rmap.ptr<Vec4b>(y);
    258         Vec4b* ryp = rmap.ptr<Vec4b>(y + 1);
    259         Vec2s* spy = spmap.ptr<Vec2s>(y);
    260         Vec2s* spyp = spmap.ptr<Vec2s>(y + 1);
    261         for (int x = 0; x < ncols - 1; ++x)
    262         {
    263             r1 = ry[x];
    264             sp1 = spy[x];
    265 
    266             r2[0] = ry[x + 1];
    267             r2[1] = ryp[x];
    268             r2[2] = ryp[x + 1];
    269             r2[3] = ryp[x];
    270 
    271             sp2[0] = spy[x + 1];
    272             sp2[1] = spyp[x];
    273             sp2[2] = spyp[x + 1];
    274             sp2[3] = spyp[x];
    275 
    276             dr[0] = dist2(r1, r2[0]);
    277             dr[1] = dist2(r1, r2[1]);
    278             dr[2] = dist2(r1, r2[2]);
    279             dsp[0] = dist2(sp1, sp2[0]);
    280             dsp[1] = dist2(sp1, sp2[1]);
    281             dsp[2] = dist2(sp1, sp2[2]);
    282 
    283             r1 = ry[x + 1];
    284             sp1 = spy[x + 1];
    285 
    286             dr[3] = dist2(r1, r2[3]);
    287             dsp[3] = dist2(sp1, sp2[3]);
    288 
    289             g.addEdge(pix(y, x, ncols), pix(y, x + 1, ncols), SegmLinkVal(dr[0], dsp[0]));
    290             g.addEdge(pix(y, x, ncols), pix(y + 1, x, ncols), SegmLinkVal(dr[1], dsp[1]));
    291             g.addEdge(pix(y, x, ncols), pix(y + 1, x + 1, ncols), SegmLinkVal(dr[2], dsp[2]));
    292             g.addEdge(pix(y, x + 1, ncols), pix(y + 1, x, ncols), SegmLinkVal(dr[3], dsp[3]));
    293         }
    294     }
    295     for (int y = 0; y < nrows - 1; ++y)
    296     {
    297         r1 = rmap.at<Vec4b>(y, ncols - 1);
    298         r2[0] = rmap.at<Vec4b>(y + 1, ncols - 1);
    299         sp1 = spmap.at<Vec2s>(y, ncols - 1);
    300         sp2[0] = spmap.at<Vec2s>(y + 1, ncols - 1);
    301         dr[0] = dist2(r1, r2[0]);
    302         dsp[0] = dist2(sp1, sp2[0]);
    303         g.addEdge(pix(y, ncols - 1, ncols), pix(y + 1, ncols - 1, ncols), SegmLinkVal(dr[0], dsp[0]));
    304     }
    305     for (int x = 0; x < ncols - 1; ++x)
    306     {
    307         r1 = rmap.at<Vec4b>(nrows - 1, x);
    308         r2[0] = rmap.at<Vec4b>(nrows - 1, x + 1);
    309         sp1 = spmap.at<Vec2s>(nrows - 1, x);
    310         sp2[0] = spmap.at<Vec2s>(nrows - 1, x + 1);
    311         dr[0] = dist2(r1, r2[0]);
    312         dsp[0] = dist2(sp1, sp2[0]);
    313         g.addEdge(pix(nrows - 1, x, ncols), pix(nrows - 1, x + 1, ncols), SegmLinkVal(dr[0], dsp[0]));
    314     }
    315 
    316     DjSets comps(g.numv);
    317 
    318     // Find adjacent components
    319     for (int v = 0; v < g.numv; ++v)
    320     {
    321         for (int e_it = g.start[v]; e_it != -1; e_it = g.edges[e_it].next)
    322         {
    323             int c1 = comps.find(v);
    324             int c2 = comps.find(g.edges[e_it].to);
    325             if (c1 != c2 && g.edges[e_it].val.dr < hr && g.edges[e_it].val.dsp < hsp)
    326                 comps.merge(c1, c2);
    327         }
    328     }
    329 
    330     std::vector<SegmLink> edges;
    331     edges.reserve(g.numv);
    332 
    333     // Prepare edges connecting differnet components
    334     for (int v = 0; v < g.numv; ++v)
    335     {
    336         int c1 = comps.find(v);
    337         for (int e_it = g.start[v]; e_it != -1; e_it = g.edges[e_it].next)
    338         {
    339             int c2 = comps.find(g.edges[e_it].to);
    340             if (c1 != c2)
    341                 edges.push_back(SegmLink(c1, c2, g.edges[e_it].val));
    342         }
    343     }
    344 
    345     // Sort all graph's edges connecting differnet components (in asceding order)
    346     std::sort(edges.begin(), edges.end());
    347 
    348     // Exclude small components (starting from the nearest couple)
    349     for (size_t i = 0; i < edges.size(); ++i)
    350     {
    351         int c1 = comps.find(edges[i].from);
    352         int c2 = comps.find(edges[i].to);
    353         if (c1 != c2 && (comps.size[c1] < minsize || comps.size[c2] < minsize))
    354             comps.merge(c1, c2);
    355     }
    356 
    357     // Compute sum of the pixel's colors which are in the same segment
    358     Mat h_src(src);
    359     std::vector<Vec4i> sumcols(nrows * ncols, Vec4i(0, 0, 0, 0));
    360     for (int y = 0; y < nrows; ++y)
    361     {
    362         Vec4b* h_srcy = h_src.ptr<Vec4b>(y);
    363         for (int x = 0; x < ncols; ++x)
    364         {
    365             int parent = comps.find(pix(y, x, ncols));
    366             Vec4b col = h_srcy[x];
    367             Vec4i& sumcol = sumcols[parent];
    368             sumcol[0] += col[0];
    369             sumcol[1] += col[1];
    370             sumcol[2] += col[2];
    371         }
    372     }
    373 
    374     // Create final image, color of each segment is the average color of its pixels
    375     _dst.create(src.size(), src.type());
    376     Mat dst = _dst.getMat();
    377 
    378     for (int y = 0; y < nrows; ++y)
    379     {
    380         Vec4b* dsty = dst.ptr<Vec4b>(y);
    381         for (int x = 0; x < ncols; ++x)
    382         {
    383             int parent = comps.find(pix(y, x, ncols));
    384             const Vec4i& sumcol = sumcols[parent];
    385             Vec4b& dstcol = dsty[x];
    386             dstcol[0] = static_cast<uchar>(sumcol[0] / comps.size[parent]);
    387             dstcol[1] = static_cast<uchar>(sumcol[1] / comps.size[parent]);
    388             dstcol[2] = static_cast<uchar>(sumcol[2] / comps.size[parent]);
    389             dstcol[3] = 255;
    390         }
    391     }
    392 }
    393 
    394 #endif // #if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
    395