CPU implementation of CLAHE

modules/imgproc/include/opencv2/imgproc/imgproc.hpp

@@ -759,6 +759,21 @@ CV_EXPORTS double compareHist( const SparseMat& H1, const SparseMat& H2, int met
 //! normalizes the grayscale image brightness and contrast by normalizing its histogram
 CV_EXPORTS_W void equalizeHist( InputArray src, OutputArray dst );
 
+class CV_EXPORTS CLAHE : public Algorithm
+{
+public:
+    virtual void apply(InputArray src, OutputArray dst) = 0;
+
+    virtual void setClipLimit(double clipLimit) = 0;
+    virtual double getClipLimit() const = 0;
+
+    virtual void setTilesGridSize(Size tileGridSize) = 0;
+    virtual Size getTilesGridSize() const = 0;
+
+    virtual void collectGarbage() = 0;
+};
+CV_EXPORTS Ptr<CLAHE> createCLAHE(double clipLimit = 40.0, Size tileGridSize = Size(8, 8));
+
 CV_EXPORTS float EMD( InputArray signature1, InputArray signature2,
                       int distType, InputArray cost=noArray(),
                       float* lowerBound=0, OutputArray flow=noArray() );

modules/imgproc/perf/perf_histogram.cpp

@@ -115,3 +115,25 @@ PERF_TEST_P(MatSize, equalizeHist,
 
     SANITY_CHECK(destination);
 }
+
+typedef tr1::tuple<Size, double> Sz_ClipLimit_t;
+typedef TestBaseWithParam<Sz_ClipLimit_t> Sz_ClipLimit;
+
+PERF_TEST_P(Sz_ClipLimit, CLAHE,
+            testing::Combine(testing::Values(::perf::szVGA, ::perf::sz720p, ::perf::sz1080p),
+                             testing::Values(0.0, 40.0))
+            )
+{
+    const Size size = get<0>(GetParam());
+    const double clipLimit = get<1>(GetParam());
+
+    Mat src(size, CV_8UC1);
+    declare.in(src, WARMUP_RNG);
+
+    Ptr<CLAHE> clahe = createCLAHE(clipLimit);
+    Mat dst;
+
+    TEST_CYCLE() clahe->apply(src, dst);
+
+    SANITY_CHECK(dst);
+}

modules/imgproc/src/histogram.cpp

@@ -3188,6 +3188,300 @@ void cv::equalizeHist( InputArray _src, OutputArray _dst )
         lutBody(heightRange);
 }
 
+// ----------------------------------------------------------------------
+// CLAHE
+
+namespace
+{
+    class CLAHE_CalcLut_Body : public cv::ParallelLoopBody
+    {
+    public:
+        CLAHE_CalcLut_Body(const cv::Mat& src, cv::Mat& lut, cv::Size tileSize, int tilesX, int tilesY, int clipLimit, float lutScale) :
+            src_(src), lut_(lut), tileSize_(tileSize), tilesX_(tilesX), tilesY_(tilesY), clipLimit_(clipLimit), lutScale_(lutScale)
+        {
+        }
+
+        void operator ()(const cv::Range& range) const;
+
+    private:
+        cv::Mat src_;
+        mutable cv::Mat lut_;
+
+        cv::Size tileSize_;
+        int tilesX_;
+        int tilesY_;
+        int clipLimit_;
+        float lutScale_;
+    };
+
+    void CLAHE_CalcLut_Body::operator ()(const cv::Range& range) const
+    {
+        const int histSize = 256;
+
+        uchar* tileLut = lut_.ptr(range.start);
+        const size_t lut_step = lut_.step;
+
+        for (int k = range.start; k < range.end; ++k, tileLut += lut_step)
+        {
+            const int ty = k / tilesX_;
+            const int tx = k % tilesX_;
+
+            // retrieve tile submatrix
+
+            cv::Rect tileROI;
+            tileROI.x = tx * tileSize_.width;
+            tileROI.y = ty * tileSize_.height;
+            tileROI.width = tileSize_.width;
+            tileROI.height = tileSize_.height;
+
+            const cv::Mat tile = src_(tileROI);
+
+            // calc histogram
+
+            int tileHist[histSize] = {0, };
+
+            int height = tileROI.height;
+            const size_t sstep = tile.step;
+            for (const uchar* ptr = tile.ptr<uchar>(0); height--; ptr += sstep)
+            {
+                int x = 0;
+                for (; x <= tileROI.width - 4; x += 4)
+                {
+                    int t0 = ptr[x], t1 = ptr[x+1];
+                    tileHist[t0]++; tileHist[t1]++;
+                    t0 = ptr[x+2]; t1 = ptr[x+3];
+                    tileHist[t0]++; tileHist[t1]++;
+                }
+
+                for (; x < tileROI.width; ++x)
+                    tileHist[ptr[x]]++;
+            }
+
+            // clip histogram
+
+            if (clipLimit_ > 0)
+            {
+                // how many pixels were clipped
+                int clipped = 0;
+                for (int i = 0; i < histSize; ++i)
+                {
+                    if (tileHist[i] > clipLimit_)
+                    {
+                        clipped += tileHist[i] - clipLimit_;
+                        tileHist[i] = clipLimit_;
+                    }
+                }
+
+                // redistribute clipped pixels
+                int redistBatch = clipped / histSize;
+                int residual = clipped - redistBatch * histSize;
+
+                for (int i = 0; i < histSize; ++i)
+                    tileHist[i] += redistBatch;
+
+                for (int i = 0; i < residual; ++i)
+                    tileHist[i]++;
+            }
+
+            // calc Lut
+
+            int sum = 0;
+            for (int i = 0; i < histSize; ++i)
+            {
+                sum += tileHist[i];
+                tileLut[i] = cv::saturate_cast<uchar>(sum * lutScale_);
+            }
+        }
+    }
+
+    class CLAHE_Interpolation_Body : public cv::ParallelLoopBody
+    {
+    public:
+        CLAHE_Interpolation_Body(const cv::Mat& src, cv::Mat& dst, const cv::Mat& lut, cv::Size tileSize, int tilesX, int tilesY) :
+            src_(src), dst_(dst), lut_(lut), tileSize_(tileSize), tilesX_(tilesX), tilesY_(tilesY)
+        {
+        }
+
+        void operator ()(const cv::Range& range) const;
+
+    private:
+        cv::Mat src_;
+        mutable cv::Mat dst_;
+        cv::Mat lut_;
+
+        cv::Size tileSize_;
+        int tilesX_;
+        int tilesY_;
+    };
+
+    void CLAHE_Interpolation_Body::operator ()(const cv::Range& range) const
+    {
+        const size_t lut_step = lut_.step;
+
+        for (int y = range.start; y < range.end; ++y)
+        {
+            const uchar* srcRow = src_.ptr<uchar>(y);
+            uchar* dstRow = dst_.ptr<uchar>(y);
+
+            const float tyf = (static_cast<float>(y) / tileSize_.height) - 0.5f;
+
+            int ty1 = cvFloor(tyf);
+            int ty2 = ty1 + 1;
+
+            const float ya = tyf - ty1;
+
+            ty1 = std::max(ty1, 0);
+            ty2 = std::min(ty2, tilesY_ - 1);
+
+            const uchar* lutPlane1 = lut_.ptr(ty1 * tilesX_);
+            const uchar* lutPlane2 = lut_.ptr(ty2 * tilesX_);
+
+            for (int x = 0; x < src_.cols; ++x)
+            {
+                const float txf = (static_cast<float>(x) / tileSize_.width) - 0.5f;
+
+                int tx1 = cvFloor(txf);
+                int tx2 = tx1 + 1;
+
+                const float xa = txf - tx1;
+
+                tx1 = std::max(tx1, 0);
+                tx2 = std::min(tx2, tilesX_ - 1);
+
+                const int srcVal = srcRow[x];
+
+                const size_t ind1 = tx1 * lut_step + srcVal;
+                const size_t ind2 = tx2 * lut_step + srcVal;
+
+                float res = 0;
+
+                res += lutPlane1[ind1] * ((1.0f - xa) * (1.0f - ya));
+                res += lutPlane1[ind2] * ((xa) * (1.0f - ya));
+                res += lutPlane2[ind1] * ((1.0f - xa) * (ya));
+                res += lutPlane2[ind2] * ((xa) * (ya));
+
+                dstRow[x] = cv::saturate_cast<uchar>(res);
+            }
+        }
+    }
+
+    class CLAHE_Impl : public cv::CLAHE
+    {
+    public:
+        CLAHE_Impl(double clipLimit = 40.0, int tilesX = 8, int tilesY = 8);
+
+        cv::AlgorithmInfo* info() const;
+
+        void apply(cv::InputArray src, cv::OutputArray dst);
+
+        void setClipLimit(double clipLimit);
+        double getClipLimit() const;
+
+        void setTilesGridSize(cv::Size tileGridSize);
+        cv::Size getTilesGridSize() const;
+
+        void collectGarbage();
+
+    private:
+        double clipLimit_;
+        int tilesX_;
+        int tilesY_;
+
+        cv::Mat srcExt_;
+        cv::Mat lut_;
+    };
+
+    CLAHE_Impl::CLAHE_Impl(double clipLimit, int tilesX, int tilesY) :
+        clipLimit_(clipLimit), tilesX_(tilesX), tilesY_(tilesY)
+    {
+    }
+
+    CV_INIT_ALGORITHM(CLAHE_Impl, "CLAHE",
+        obj.info()->addParam(obj, "clipLimit", obj.clipLimit_);
+        obj.info()->addParam(obj, "tilesX", obj.tilesX_);
+        obj.info()->addParam(obj, "tilesY", obj.tilesY_))
+
+    void CLAHE_Impl::apply(cv::InputArray _src, cv::OutputArray _dst)
+    {
+        cv::Mat src = _src.getMat();
+
+        CV_Assert( src.type() == CV_8UC1 );
+
+        _dst.create( src.size(), src.type() );
+        cv::Mat dst = _dst.getMat();
+
+        const int histSize = 256;
+
+        lut_.create(tilesX_ * tilesY_, histSize, CV_8UC1);
+
+        cv::Size tileSize;
+        cv::Mat srcForLut;
+
+        if (src.cols % tilesX_ == 0 && src.rows % tilesY_ == 0)
+        {
+            tileSize = cv::Size(src.cols / tilesX_, src.rows / tilesY_);
+            srcForLut = src;
+        }
+        else
+        {
+            cv::copyMakeBorder(src, srcExt_, 0, tilesY_ - (src.rows % tilesY_), 0, tilesX_ - (src.cols % tilesX_), cv::BORDER_REFLECT_101);
+
+            tileSize = cv::Size(srcExt_.cols / tilesX_, srcExt_.rows / tilesY_);
+            srcForLut = srcExt_;
+        }
+
+        const int tileSizeTotal = tileSize.area();
+        const float lutScale = static_cast<float>(histSize - 1) / tileSizeTotal;
+
+        int clipLimit = 0;
+        if (clipLimit_ > 0.0)
+        {
+            clipLimit = static_cast<int>(clipLimit_ * tileSizeTotal / histSize);
+            clipLimit = std::max(clipLimit, 1);
+        }
+
+        CLAHE_CalcLut_Body calcLutBody(srcForLut, lut_, tileSize, tilesX_, tilesY_, clipLimit, lutScale);
+        cv::parallel_for_(cv::Range(0, tilesX_ * tilesY_), calcLutBody);
+
+        CLAHE_Interpolation_Body interpolationBody(src, dst, lut_, tileSize, tilesX_, tilesY_);
+        cv::parallel_for_(cv::Range(0, src.rows), interpolationBody);
+    }
+
+    void CLAHE_Impl::setClipLimit(double clipLimit)
+    {
+        clipLimit_ = clipLimit;
+    }
+
+    double CLAHE_Impl::getClipLimit() const
+    {
+        return clipLimit_;
+    }
+
+    void CLAHE_Impl::setTilesGridSize(cv::Size tileGridSize)
+    {
+        tilesX_ = tileGridSize.width;
+        tilesY_ = tileGridSize.height;
+    }
+
+    cv::Size CLAHE_Impl::getTilesGridSize() const
+    {
+        return cv::Size(tilesX_, tilesY_);
+    }
+
+    void CLAHE_Impl::collectGarbage()
+    {
+        srcExt_.release();
+        lut_.release();
+    }
+}
+
+cv::Ptr<cv::CLAHE> cv::createCLAHE(double clipLimit, cv::Size tileGridSize)
+{
+    return new CLAHE_Impl(clipLimit, tileGridSize.width, tileGridSize.height);
+}
+
+// ----------------------------------------------------------------------
+
 /* Implementation of RTTI and Generic Functions for CvHistogram */
 #define CV_TYPE_NAME_HIST "opencv-hist"
 
@@ -3339,4 +3633,3 @@ CvType hist_type( CV_TYPE_NAME_HIST, icvIsHist, (CvReleaseFunc)cvReleaseHist,
                   icvReadHist, icvWriteHist, (CvCloneFunc)icvCloneHist );
 
 /* End of file. */
-