形态学滤波

怼烎@ 2022-06-17 07:29 232阅读 0赞

**基本形态学滤波：膨胀，腐蚀**

**膨胀腐蚀功能：消除噪声，分割出独立的图像元素，在图像中连接相邻的元素，寻找图像中的明显的极大值区域或极小值区域，求出图像的梯度。**

**膨胀是向着高亮部分进行膨胀，腐蚀亮度腐蚀，呈现暗区域。**

**膨胀：求局部最大值，就是将图像（图像某一区域A）与核（B）进行卷积，有有一参考点（锚点）。核B与图形卷积，即计算核B覆盖的区域的像素点最大值，并把这个最大值赋给参考点指定的像素。**

**腐蚀是与膨胀相反的过程。**

**相关源码：**

**腐蚀：**

void cv::erode( InputArray src, OutputArray dst, InputArray kernel,
                    Point anchor, int iterations,
                    int borderType, const Scalar& borderValue )
    {
        morphOp( MORPH_ERODE, src, dst, kernel, anchor, iterations, borderType, borderValue );
    }

static void morphOp( int op, InputArray _src, OutputArray _dst,
                         InputArray _kernel,
                         Point anchor, int iterations,
                         int borderType, const Scalar& borderValue )
    {
        Mat kernel = _kernel.getMat();
        Size ksize = !kernel.empty() ? kernel.size() : Size(3,3);
        anchor = normalizeAnchor(anchor, ksize);
    
        CV_OCL_RUN(_dst.isUMat() && _src.dims() <= 2 && _src.channels() <= 4 &&
                   borderType == cv::BORDER_CONSTANT && borderValue == morphologyDefaultBorderValue() &&
                   (op == MORPH_ERODE || op == MORPH_DILATE) &&
                   anchor.x == ksize.width >> 1 && anchor.y == ksize.height >> 1,
                   ocl_morphOp(_src, _dst, kernel, anchor, iterations, op, borderType, borderValue) )
    
        if (iterations == 0 || kernel.rows*kernel.cols == 1)
        {
            _src.copyTo(_dst);
            return;
        }
    
        if (kernel.empty())
        {
            kernel = getStructuringElement(MORPH_RECT, Size(1+iterations*2,1+iterations*2));
            anchor = Point(iterations, iterations);
            iterations = 1;
        }
        else if( iterations > 1 && countNonZero(kernel) == kernel.rows*kernel.cols )
        {
            anchor = Point(anchor.x*iterations, anchor.y*iterations);
            kernel = getStructuringElement(MORPH_RECT,
                                           Size(ksize.width + (iterations-1)*(ksize.width-1),
                                                ksize.height + (iterations-1)*(ksize.height-1)),
                                           anchor);
            iterations = 1;
        }
    
    #if IPP_VERSION_X100 >= 801
        CV_IPP_CHECK()
        {
            if( IPPMorphOp(op, _src, _dst, kernel, anchor, iterations, borderType, borderValue) )
            {
                CV_IMPL_ADD(CV_IMPL_IPP);
                return;
            }
        }
    #endif
    
        Mat src = _src.getMat();
        _dst.create( src.size(), src.type() );
        Mat dst = _dst.getMat();
    
        int nStripes = 1;
    #if defined HAVE_TEGRA_OPTIMIZATION
        if (src.data != dst.data && iterations == 1 &&  //NOTE: threads are not used for inplace processing
            (borderType & BORDER_ISOLATED) == 0 && //TODO: check border types
            src.rows >= 64 ) //NOTE: just heuristics
            nStripes = 4;
    #endif
    
        parallel_for_(Range(0, nStripes),
                      MorphologyRunner(src, dst, nStripes, iterations, op, kernel, anchor, borderType, borderType, borderValue));
    }

膨胀（dilate）：

void cv::dilate( InputArray src, OutputArray dst, InputArray kernel,
                     Point anchor, int iterations,
                     int borderType, const Scalar& borderValue )
    {
        morphOp( MORPH_DILATE, src, dst, kernel, anchor, iterations, borderType, borderValue );
    }

高级形态学滤波：开运算、闭运算、形态学梯度、顶帽、黑帽

**1、开运算**

**     先腐蚀后膨胀，可用来消除小物体，在纤细点处分离物体，并且在平滑较大物体的边界的同时不明显改变其面积，但放大了裂缝或者局部低亮度区域。**

**2、闭运算**

**    先膨胀后腐蚀，排除小型黑洞（黑色区域）。**

**3、形态学梯度**

**    膨胀图与腐蚀图之差，保留物体的边缘轮廓。**

**4、顶帽**

**    原图像与开运算的结果图之差。得到的效果图突出了比原图轮廓周围的区域更明亮的区域，且与选择核的大小相关。顶帽运算往往用来分离比邻近点亮一些的斑点。在一幅图像具有大幅的背景而微小物品比较有规律的情况下，可用其进行背景提取。**

**5、黑帽**

**    闭运算与原图像之差，突出了比原图像轮廓周围的更暗区域，与选择核有关。黑帽运算用来分离比邻近点暗一些的斑点，有着非常完美的轮廓。**

**源码：**

void cv::morphologyEx( InputArray _src, OutputArray _dst, int op, InputArray _kernel, Point anchor, int iterations, int borderType, const Scalar& borderValue ) { Mat kernel = _kernel.getMat(); if (kernel.empty()) { kernel = getStructuringElement(MORPH_RECT, Size(3,3), Point(1,1)); } #ifdef HAVE_OPENCL Size ksize = kernel.size(); anchor = normalizeAnchor(anchor, ksize); CV_OCL_RUN(_dst.isUMat() && _src.dims() <= 2 && _src.channels() <= 4 && anchor.x == ksize.width >> 1 && anchor.y == ksize.height >> 1 && borderType == cv::BORDER_CONSTANT && borderValue == morphologyDefaultBorderValue(), ocl_morphologyEx(_src, _dst, op, kernel, anchor, iterations, borderType, borderValue)) #endif Mat src = _src.getMat(), temp; _dst.create(src.size(), src.type()); Mat dst = _dst.getMat(); switch( op ) { case MORPH_ERODE: erode( src, dst, kernel, anchor, iterations, borderType, borderValue ); break; case MORPH_DILATE: dilate( src, dst, kernel, anchor, iterations, borderType, borderValue ); break; case MORPH_OPEN: erode( src, dst, kernel, anchor, iterations, borderType, borderValue ); dilate( dst, dst, kernel, anchor, iterations, borderType, borderValue ); break; case CV_MOP_CLOSE: dilate( src, dst, kernel, anchor, iterations, borderType, borderValue ); erode( dst, dst, kernel, anchor, iterations, borderType, borderValue ); break; case CV_MOP_GRADIENT: erode( src, temp, kernel, anchor, iterations, borderType, borderValue ); dilate( src, dst, kernel, anchor, iterations, borderType, borderValue ); dst -= temp; break; case CV_MOP_TOPHAT: if( src.data != dst.data ) temp = dst; erode( src, temp, kernel, anchor, iterations, borderType, borderValue ); dilate( temp, temp, kernel, anchor, iterations, borderType, borderValue ); dst = src - temp; break; case CV_MOP_BLACKHAT: if( src.data != dst.data ) temp = dst; dilate( src, temp, kernel, anchor, iterations, borderType, borderValue ); erode( temp, temp, kernel, anchor, iterations, borderType, borderValue ); dst = temp - src; break; default: CV_Error( CV_StsBadArg, "unknown morphological operation" ); } }