Bernsen算法是比较出名的二值化算法，网上很多Bernsen代码是基于Matlab的，本人觉得其速度比较慢，所以便基于OpenCV改写了其算法，具体参考的博客链接已经忘记了，希望博主原谅。如果缺少某些函数，比如最大值最小值函数，可以参考本人其他博客，里面会提供。废话不多说，直接上代码：
/** @brief 得到矩阵中的最大值与最小值 @param m 单通道CV_8UC1类型矩阵 @param maxValue 最大值 @param minValue 最小值 */ static void GetMatMaxMin(const cv::Mat& m, int& maxValue, int& minValue) { CV_Assert(m.type() == CV_8UC1); maxValue = INT_MIN; minValue = INT_MAX; for (int y = 0; y < m.rows; ++y) { for (int x = 0; x < m.cols; ++x) { int v = m.at<uchar>(y, x); if (v > maxValue) maxValue = v; if (v < minValue) minValue = v; } } } void AutoThresholder::Bernsen(const cv::Mat & src, cv::Mat & dst, cv::Size wndSize) { CV_Assert(src.type() == CV_8UC1); CV_Assert((wndSize.width % 2 == 1) && (wndSize.height % 2 == 1)); CV_Assert((wndSize.width <= src.cols) && (wndSize.height <= src.rows)); cv::Mat meanMat = cv::Mat::zeros(src.rows, src.cols, CV_8UC1); for (int y = wndSize.height / 2; y <= src.rows - wndSize.height / 2 - 1; ++y) { for (int x = wndSize.width / 2; x <= src.cols - wndSize.width / 2 - 1; ++x) { int value = src.at<uchar>(y, x); cv::Point center = cv::Point(x, y); cv::Point topLeftPoint = cv::Point(x - wndSize.width / 2, y - wndSize.height / 2); cv::Rect wnd = cv::Rect(topLeftPoint.x, topLeftPoint.y, wndSize.width, wndSize.height); int maxValue = 0; int minValue = 0; cv::Mat roiMat = src(wnd); GetMatMaxMin(roiMat, maxValue, minValue); int meanValue = (maxValue + minValue) / 2.0; meanMat.at<uchar>(y, x) = meanValue; } } // 阈值分割 dst = cv::Mat::zeros(src.rows, src.cols, CV_8UC1); for (int y = 0; y < src.rows; ++y) { for (int x = 0; x < src.cols; ++x) { int value = src.at<uchar>(y, x); int meanValue = meanMat.at<uchar>(y, x); if (value > meanValue) { dst.at<uchar>(y, x) = 255; } else { dst.at<uchar>(y, x) = 0; } } } } void AutoThresholder::Bernsen(const cv::Mat & src, cv::Mat & dst, cv::Size wndSize, int differMax, int meanMax) { CV_Assert(src.type() == CV_8UC1); CV_Assert((wndSize.width % 2 == 1) && (wndSize.height % 2 == 1)); CV_Assert((wndSize.width <= src.cols) && (wndSize.height <= src.rows)); // 计算均值矩阵和差异矩阵 cv::Mat meanMat = cv::Mat::zeros(src.rows, src.cols, CV_8UC1); cv::Mat differMat = cv::Mat::zeros(src.rows, src.cols, CV_8UC1); for (int y = wndSize.height / 2; y <= src.rows - wndSize.height / 2 - 1; ++y) { for (int x = wndSize.width / 2; x <= src.cols - wndSize.width / 2 - 1; ++x) { int value = src.at<uchar>(y, x); cv::Point center = cv::Point(x, y); cv::Point topLeftPoint = cv::Point(x - wndSize.width / 2, y - wndSize.height / 2); cv::Rect wnd = cv::Rect(topLeftPoint.x, topLeftPoint.y, wndSize.width, wndSize.height); int maxValue = 0; int minValue = 0; cv::Mat roiMat = src(wnd); GetMatMaxMin(roiMat, maxValue, minValue); int meanValue = (maxValue + minValue) / 2.0; int differValue = maxValue - minValue; meanMat.at<uchar>(y, x) = meanValue; differMat.at<uchar>(y, x) = differValue; } } // 赋值 dst = cv::Mat::zeros(src.rows, src.cols, CV_8UC1); for (int y = 0; y < differMat.rows; ++y) { for (int x = 0; x < differMat.cols; ++x) { int differValue = differMat.at<uchar>(y, x); if (differValue > differMax) { // blog写的很迷糊, 直说meanValue是阈值 // 本人认为是边界部分,可以是0,也可以是255 dst.at<uchar>(y, x) = 255; } else if (differValue < differMax) { int meanValue = meanMat.at<uchar>(y, x); if (meanValue > meanMax) { dst.at<uchar>(y, x) = 255; } else { dst.at<uchar>(y, x) = 0; } } else { // TODO dst.at<uchar>(y, x) = 0; } } } }文章来源: 基于opencv的Bernsen二值化算法