当前位置：首页 > news >正文

C++基于opencv的视频质量检测--图像清晰度检测

news 2025/8/27 0:10:21

文章目录

- 0.引言
- 1. 原始算法实现
- 2. 优化思路
- 3. 优化后的代码
- 4. 代码详细解读

0.引言

视频质量图像清晰度检测已在C++基于opencv4的视频质量检测中有所介绍，本文将详细介绍其优化版本。

1. 原始算法实现

原始代码：

double sharpnessDetect(const cv::Mat& srcImg) {const int kGaussianSize = 3;  // 高斯模糊的核大小cv::Mat grayImg;if (srcImg.channels() != 1) {cv::cvtColor(srcImg, grayImg, cv::COLOR_BGR2GRAY);} else {grayImg = srcImg;}cv::Mat blurredImg;cv::GaussianBlur(grayImg, blurredImg, cv::Size(kGaussianSize, kGaussianSize), 0);uint64_t sumFver = 0;uint64_t sumFhor = 0;uint64_t sumVver = 0;uint64_t sumVhor = 0;double blurFactor = 0.0;for (int r = 0; r < grayImg.rows; ++r) {for (int c = 0; c < grayImg.cols; ++c) {uint64_t diffFver = 0;uint64_t diffFhor = 0;uint64_t diffBver = 0;uint64_t diffBhor = 0;if (r != 0) {diffFver = static_cast<uint64_t>(std::abs(grayImg.at<uchar>(r, c) - grayImg.at<uchar>(r - 1, c)));}if (c != 0) {diffFhor = static_cast<uint64_t>(std::abs(grayImg.at<uchar>(r, c) - grayImg.at<uchar>(r, c - 1)));}if (r != 0) {diffBver = static_cast<uint64_t>(std::abs(blurredImg.at<uchar>(r, c) - blurredImg.at<uchar>(r - 1, c)));}if (c != 0) {diffBhor = static_cast<uint64_t>(std::abs(blurredImg.at<uchar>(r, c) - blurredImg.at<uchar>(r, c - 1)));}uint64_t verDiff = (diffFver > diffBver) ? (diffFver - diffBver) : 0;uint64_t horDiff = (diffFhor > diffBhor) ? (diffFhor - diffBhor) : 0;sumFver += diffFver;sumFhor += diffFhor;sumVver += verDiff;sumVhor += horDiff;}}double bFver = (static_cast<double>(sumFver - sumVver)) / (static_cast<double>(sumFver) + 1.0);double bFhor = (static_cast<double>(sumFhor - sumVhor)) / (static_cast<double>(sumFhor) + 1.0);blurFactor = (bFver > bFhor) ? bFver : bFhor;return 1.0 - blurFactor;
}

以下是原始代码的主要步骤：

图像预处理：将输入图像转换为灰度图。
高斯模糊：对灰度图像进行高斯模糊处理，得到模糊图像。
梯度计算：通过遍历每个像素，计算原始图像和模糊图像在垂直和水平方向上的梯度差异。
模糊因子计算：根据梯度差异，计算模糊因子，进而评估图像的清晰度。

存在的问题：

效率低下：使用嵌套的for循环遍历每个像素，手动计算梯度，处理大尺寸图像时效率较低。

2. 优化思路

为了解决上述问题，我们对原始算法进行了以下优化：

向量化操作，避免显式循环：利用OpenCV的矩阵运算和函数，对整个图像进行批量处理，提高计算效率。
使用Sobel算子计算梯度：Sobel算子是常用的梯度计算方法，能够有效地提取图像的边缘信息。

3. 优化后的代码

以下是优化后的sharpnessDetect函数：

double sharpnessDetect(const cv::Mat& srcImg) {const int kGaussianSize = 3;  // 高斯模糊的核大小cv::Mat grayImg;if (srcImg.channels() != 1) {cv::cvtColor(srcImg, grayImg, cv::COLOR_BGR2GRAY);} else {grayImg = srcImg;}// 对灰度图像进行高斯模糊cv::Mat blurredImg;cv::GaussianBlur(grayImg, blurredImg, cv::Size(kGaussianSize, kGaussianSize), 0);// 计算原始图像和模糊图像的梯度cv::Mat gradXOrig, gradYOrig;cv::Mat gradXBlur, gradYBlur;cv::Sobel(grayImg, gradXOrig, CV_64F, 1, 0, 3);cv::Sobel(grayImg, gradYOrig, CV_64F, 0, 1, 3);cv::Sobel(blurredImg, gradXBlur, CV_64F, 1, 0, 3);cv::Sobel(blurredImg, gradYBlur, CV_64F, 0, 1, 3);// 计算梯度的绝对值cv::Mat absGradXOrig = cv::abs(gradXOrig);cv::Mat absGradYOrig = cv::abs(gradYOrig);cv::Mat absGradXBlur = cv::abs(gradXBlur);cv::Mat absGradYBlur = cv::abs(gradYBlur);// 计算梯度差的正值部分cv::Mat diffX = absGradXOrig - absGradXBlur;cv::Mat diffY = absGradYOrig - absGradYBlur;diffX = cv::max(diffX, 0);diffY = cv::max(diffY, 0);// 计算梯度的总和double sumFhor = cv::sum(absGradXOrig)[0];double sumFver = cv::sum(absGradYOrig)[0];// 计算梯度差的总和double sumVhor = cv::sum(diffX)[0];double sumVver = cv::sum(diffY)[0];// 计算模糊因子double bFver = (sumFver - sumVver) / (sumFver + 1e-6);double bFhor = (sumFhor - sumVhor) / (sumFhor + 1e-6);double blurFactor = std::max(bFver, bFhor);// 返回清晰度得分return 1.0 - blurFactor;
}