Merge pull request #29434 from Prasadayus:canny_ipp_extract

Extracting IPP integaration as HAL for Canny #29434

Backport of : https://github.com/opencv/opencv/pull/29433

**Performance Numbers on Intel(R) Core(TM) i9-11900K:** https://docs.google.com/spreadsheets/d/1RMvUZP1tSQtdjuNQY9JasYmlx1L5iN9XWGhXtG5u1uI/edit?usp=sharing 

### Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
      Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake

hal/ipp/CMakeLists.txt

@@ -23,6 +23,7 @@ add_library(ipphal STATIC
     "${CMAKE_CURRENT_SOURCE_DIR}/src/color_ipp.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/src/filter_ipp.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/src/matchtemplate_ipp.cpp"
+    "${CMAKE_CURRENT_SOURCE_DIR}/src/canny_ipp.cpp"
 )
 
 #TODO: HAVE_IPP_ICV and HAVE_IPP_IW added as private macro till OpenCV itself is

hal/ipp/include/ipp_hal_imgproc.hpp

@@ -146,4 +146,20 @@ int ipp_hal_matchTemplate(const uchar* src_data, size_t src_step, int src_width,
 
 #endif // IPP_VERSION_X100 >= 700
 
+#define IPP_DISABLE_PERF_CANNY_MT 1 // cv::Canny OpenCV MT performance is better
+
+#if defined(HAVE_IPP_IW)
+int ipp_hal_canny(const uchar* src_data, size_t src_step, uchar* dst_data, size_t dst_step,
+                  int width, int height, int cn,
+                  double lowThreshold, double highThreshold, int ksize, bool L2gradient);
+#undef cv_hal_canny
+#define cv_hal_canny ipp_hal_canny
+
+int ipp_hal_canny_deriv(const short* dx_data, size_t dx_step, const short* dy_data, size_t dy_step,
+                        uchar* dst_data, size_t dst_step, int width, int height, int cn,
+                        double lowThreshold, double highThreshold, bool L2gradient);
+#undef cv_hal_canny_deriv
+#define cv_hal_canny_deriv ipp_hal_canny_deriv
+#endif
+
 #endif //__IPP_HAL_IMGPROC_HPP__

hal/ipp/src/canny_ipp.cpp

@@ -0,0 +1,95 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html
+// Copyright (C) 2026, BigVision LLC, all rights reserved.
+// Third party copyrights are property of their respective owners.
+
+#include "ipp_hal_imgproc.hpp"
+
+#ifdef HAVE_IPP_IW
+
+#include <opencv2/core.hpp>
+#include "precomp_ipp.hpp"
+
+int ipp_hal_canny(const uchar* src_data, size_t src_step, uchar* dst_data, size_t dst_step,
+                  int width, int height, int cn,
+                  double lowThreshold, double highThreshold, int ksize, bool L2gradient)
+{
+    CV_HAL_CHECK_USE_IPP();
+
+#if IPP_DISABLE_PERF_CANNY_MT
+    if(cv::getNumThreads() > 1)
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+#endif
+
+    if(width <= 3 || height <= 3)
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+
+    if(cn != 1)
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+
+    IppiMaskSize kernel;
+    if(ksize == 3)
+        kernel = ippMskSize3x3;
+    else if(ksize == 5)
+        kernel = ippMskSize5x5;
+    else
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+
+    IppNormType norm = L2gradient ? ippNormL2 : ippNormL1;
+
+    try
+    {
+        ::ipp::IwiImage iwSrc, iwDst;
+        iwSrc.Init(IwiSize{width, height}, ipp8u, cn, ::ipp::IwiBorderSize(), (void*)src_data, IwSize(src_step));
+        iwDst.Init(IwiSize{width, height}, ipp8u, cn, ::ipp::IwiBorderSize(), dst_data, IwSize(dst_step));
+
+        CV_INSTRUMENT_FUN_IPP(::ipp::iwiFilterCanny, iwSrc, iwDst, (float)lowThreshold, (float)highThreshold,
+                              ::ipp::IwiFilterCannyParams(ippFilterSobel, kernel, norm), ippBorderRepl);
+    }
+    catch (const ::ipp::IwException &)
+    {
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+    }
+
+    return CV_HAL_ERROR_OK;
+}
+
+int ipp_hal_canny_deriv(const short* dx_data, size_t dx_step, const short* dy_data, size_t dy_step,
+                        uchar* dst_data, size_t dst_step, int width, int height, int cn,
+                        double lowThreshold, double highThreshold, bool L2gradient)
+{
+    CV_HAL_CHECK_USE_IPP();
+
+#if IPP_DISABLE_PERF_CANNY_MT
+    if(cv::getNumThreads() > 1)
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+#endif
+
+    if(width <= 3 || height <= 3)
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+
+    if(cn != 1)
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+
+    IppNormType norm = L2gradient ? ippNormL2 : ippNormL1;
+
+    try
+    {
+        ::ipp::IwiImage iwSrcDx, iwSrcDy, iwDst;
+        iwSrcDx.Init(IwiSize{width, height}, ipp16s, cn, ::ipp::IwiBorderSize(), (void*)dx_data, IwSize(dx_step));
+        iwSrcDy.Init(IwiSize{width, height}, ipp16s, cn, ::ipp::IwiBorderSize(), (void*)dy_data, IwSize(dy_step));
+        iwDst.Init(IwiSize{width, height}, ipp8u, cn, ::ipp::IwiBorderSize(), dst_data, IwSize(dst_step));
+
+        CV_INSTRUMENT_FUN_IPP(::ipp::iwiFilterCannyDeriv, iwSrcDx, iwSrcDy, iwDst, (float)lowThreshold, (float)highThreshold,
+                              ::ipp::IwiFilterCannyDerivParams(norm));
+    }
+    catch (const ::ipp::IwException &)
+    {
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+    }
+
+    return CV_HAL_ERROR_OK;
+}
+
+#endif

modules/core/include/opencv2/core/private.hpp

@@ -211,7 +211,6 @@ T* allocSingletonNew() { return new(allocSingletonNewBuffer(sizeof(T))) T(); }
 // Temporary disabled named IPP region. Performance
 #define IPP_DISABLE_PERF_COPYMAKE       1 // performance variations
 #define IPP_DISABLE_PERF_TRUE_DIST_MT   1 // cv::distanceTransform OpenCV MT performance is better
-#define IPP_DISABLE_PERF_CANNY_MT       1 // cv::Canny OpenCV MT performance is better
 
 #ifdef HAVE_IPP
 #include "ippversion.h"

modules/imgproc/src/canny.cpp

@@ -48,85 +48,6 @@
 namespace cv
 {
 
-#ifdef HAVE_IPP
-static bool ipp_Canny(const Mat& src , const Mat& dx_, const Mat& dy_, Mat& dst, float low,  float high, bool L2gradient, int aperture_size)
-{
-#ifdef HAVE_IPP_IW
-    CV_INSTRUMENT_REGION_IPP();
-
-#if IPP_DISABLE_PERF_CANNY_MT
-    if(cv::getNumThreads()>1)
-        return false;
-#endif
-
-    ::ipp::IwiSize size(dst.cols, dst.rows);
-    IppDataType    type     = ippiGetDataType(dst.depth());
-    int            channels = dst.channels();
-    IppNormType    norm     = (L2gradient)?ippNormL2:ippNormL1;
-
-    if(size.width <= 3 || size.height <= 3)
-        return false;
-
-    if(channels != 1)
-        return false;
-
-    if(type != ipp8u)
-        return false;
-
-    if(src.empty())
-    {
-        try
-        {
-            ::ipp::IwiImage iwSrcDx;
-            ::ipp::IwiImage iwSrcDy;
-            ::ipp::IwiImage iwDst;
-
-            ippiGetImage(dx_, iwSrcDx);
-            ippiGetImage(dy_, iwSrcDy);
-            ippiGetImage(dst, iwDst);
-
-            CV_INSTRUMENT_FUN_IPP(::ipp::iwiFilterCannyDeriv, iwSrcDx, iwSrcDy, iwDst, low, high, ::ipp::IwiFilterCannyDerivParams(norm));
-        }
-        catch (const ::ipp::IwException &)
-        {
-            return false;
-        }
-    }
-    else
-    {
-        IppiMaskSize kernel;
-
-        if(aperture_size == 3)
-            kernel = ippMskSize3x3;
-        else if(aperture_size == 5)
-            kernel = ippMskSize5x5;
-        else
-            return false;
-
-        try
-        {
-            ::ipp::IwiImage iwSrc;
-            ::ipp::IwiImage iwDst;
-
-            ippiGetImage(src, iwSrc);
-            ippiGetImage(dst, iwDst);
-
-            CV_INSTRUMENT_FUN_IPP(::ipp::iwiFilterCanny, iwSrc, iwDst, low, high, ::ipp::IwiFilterCannyParams(ippFilterSobel, kernel, norm), ippBorderRepl);
-        }
-        catch (const ::ipp::IwException &)
-        {
-            return false;
-        }
-    }
-
-    return true;
-#else
-    CV_UNUSED(src); CV_UNUSED(dx_); CV_UNUSED(dy_); CV_UNUSED(dst); CV_UNUSED(low); CV_UNUSED(high); CV_UNUSED(L2gradient); CV_UNUSED(aperture_size);
-    return false;
-#endif
-}
-#endif
-
 #ifdef HAVE_OPENCL
 
 template <bool useCustomDeriv>
@@ -803,8 +724,6 @@ void Canny( InputArray _src, OutputArray _dst,
     CALL_HAL(canny, cv_hal_canny, src.data, src.step, dst.data, dst.step, src.cols, src.rows, src.channels(),
              low_thresh, high_thresh, aperture_size, L2gradient);
 
-    CV_IPP_RUN_FAST(ipp_Canny(src, Mat(), Mat(), dst, (float)low_thresh, (float)high_thresh, L2gradient, aperture_size))
-
     if (L2gradient)
     {
         low_thresh = std::min(32767.0, low_thresh);
@@ -879,7 +798,8 @@ void Canny( InputArray _dx, InputArray _dy, OutputArray _dst,
     Mat dx = _dx.getMat();
     Mat dy = _dy.getMat();
 
-    CV_IPP_RUN_FAST(ipp_Canny(Mat(), dx, dy, dst, (float)low_thresh, (float)high_thresh, L2gradient, 0))
+    CALL_HAL(canny_deriv, cv_hal_canny_deriv, dx.ptr<short>(), dx.step, dy.ptr<short>(), dy.step,
+             dst.data, dst.step, dx.cols, dx.rows, dx.channels(), low_thresh, high_thresh, L2gradient);
 
     if (L2gradient)
     {

modules/imgproc/src/hal_replacement.hpp

@@ -1520,6 +1520,27 @@ inline int hal_ni_canny(const uchar* src_data, size_t src_step, uchar* dst_data,
 #define cv_hal_canny hal_ni_canny
 //! @endcond
 
+/**
+   @brief Canny edge detector from image derivatives
+   @param dx_data Source image x-derivative data
+   @param dx_step Source image x-derivative step
+   @param dy_data Source image y-derivative data
+   @param dy_step Source image y-derivative step
+   @param dst_data Destination image data
+   @param dst_step Destination image step
+   @param width Source image width
+   @param height Source image height
+   @param cn Number of channels
+   @param lowThreshold low hresholds value
+   @param highThreshold high thresholds value
+   @param L2gradient Flag, indicating use L2 or L1 norma.
+*/
+inline int hal_ni_canny_deriv(const short* dx_data, size_t dx_step, const short* dy_data, size_t dy_step, uchar* dst_data, size_t dst_step, int width, int height, int cn, double lowThreshold, double highThreshold, bool L2gradient) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
+
+//! @cond IGNORED
+#define cv_hal_canny_deriv hal_ni_canny_deriv
+//! @endcond
+
 /**
    @brief Calculates all of the moments up to the third order of a polygon or rasterized shape for image
    @param src_data Source image data