#include #include #include // 读取二值化的单通道一位图片并生成掩膜 std::vector> generateMaskFromImage(const std::string& imagePath, int widthBlocks, int heightBlocks, int threshold = 10) { // 读取图像 cv::Mat image = cv::imread(imagePath, cv::IMREAD_GRAYSCALE); // 检查图像是否成功读取 if (image.empty()) { std::cerr << "无法加载图像,请检查路径是否正确: " << imagePath << std::endl; exit(EXIT_FAILURE); } // 确保图像是二值化的 cv::threshold(image, image, 128, 255, cv::THRESH_BINARY); // 获取图像的宽度和高度 int imageWidth = image.cols; int imageHeight = image.rows; // 计算每个块的宽度和高度 int blockWidth = imageWidth / widthBlocks; int blockHeight = imageHeight / heightBlocks; // 创建掩膜矩阵 std::vector> mask(heightBlocks, std::vector(widthBlocks, false)); // 遍历每个块并统计白色像素点的数量 for (int i = 0; i < heightBlocks; ++i) { for (int j = 0; j < widthBlocks; ++j) { // 计算块的起始和结束位置 int x_start = j * blockWidth; int y_start = i * blockHeight; int x_end = (j == widthBlocks - 1) ? imageWidth : (j + 1) * blockWidth; int y_end = (i == heightBlocks - 1) ? imageHeight : (i + 1) * blockHeight; // 提取当前块 cv::Mat block = image(cv::Rect(x_start, y_start, x_end - x_start, y_end - y_start)); // 统计块中白色像素的数量 int whitePixelCount = cv::countNonZero(block); // 如果白色像素数大于阈值,将该块标记为 true if (whitePixelCount > threshold) { mask[i][j] = true; } } } return mask; } int main() { // 指定图像路径 std::string imagePath = "C:\\Users\\zjc\\Desktop\\diguandai.png"; // 设置分块数量和白色像素点阈值 int widthBlocks = 24; int heightBlocks = 24; int threshold = 20; // 生成掩膜 std::vector> mask = generateMaskFromImage(imagePath, widthBlocks, heightBlocks, threshold); // 打印掩膜结果 for (int i = 0; i < heightBlocks; ++i) { for (int j = 0; j < widthBlocks; ++j) { std::cout << (mask[i][j] ? "1 " : "0 "); } std::cout << std::endl; } return 0; }