实验三曝光融合

阅读：评论：0

实验三曝光融合

3.1 任务简介

当拍摄真实场景照片时，有时会出现光照不均匀情况：亮区域会出现过曝光，而暗区域会欠曝光，因此无法通过一次曝光获取到所有图像细节。所以可以应用多次不同曝光值下所拍摄的照片进行曝光融合，HDR就是一种常用到曝光融合算法的拍摄技术。

3.2 算法实现

 //通过灰度均值获取权值float weight[16];float sum=0;float sum2=0;float yu[16];for (i = 0; i < 16; i++){yu[i] = mean_gray[i] - 128;if (yu[i] < 0){yu[i] = - yu[i];}sum += yu[i]; }for (i = 0; i < 16; i++){sum2 += sum / yu[i];}for (i = 0; i < 16; i++){weight[i] =( sum / yu[i] )/sum2;cout << "权值" << weight[i];}

既然谈到了灰度，也来说一下我是如何获取灰度的吧，在OpenCV库中，有一个名为meanStdDev()的函数，它的作用是计算矩阵的均值和标准偏差，由于在本次实验中使用的是彩图，所以会返回一个三通道的数组，只要将他们加和之后再取平均即可得到灰度平均值。代码的话我也贴在下方。

	Scalar     mean;Scalar     stddev;float mean_gray[16];  //灰度均值//获取灰度均值for (i = 0; i < 16; i++){meanStdDev(image[i], mean, stddev);  //分别获取图的均值与方差，在本算法中仅使用了均值mean_gray[i] = (mean[0] + mean[1] + mean[2])/3;cout << "<----mean_pxl------>" << mean_gray[i] << "<-------stddev_pxl------->" << stddev[0] << endl;}

整体的代码的话就如下图了，我在读图片的时候用了比较笨的办法，这里是可以优化的。


```cpp
int main(int argc, char** argv)
{int i=0;Mat image[16];//定义一个图像数组用于存放待处理图像image[0] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0061.jpg", 1); // Read the fileimage[1] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0062.jpg", 1); // Read the fileimage[2] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0063.jpg", 1); // Read the fileimage[3] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0064.jpg", 1); // Read the fileimage[4] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0065.jpg", 1); // Read the fileimage[5] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0066.jpg", 1); // Read the fileimage[6] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0067.jpg", 1); // Read the fileimage[7] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0068.jpg", 1); // Read the fileimage[8] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0069.jpg", 1); // Read the fileimage[9] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0070.jpg", 1); // Read the fileimage[10] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0071.jpg", 1); // Read the fileimage[11] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0071.jpg", 1); // Read the fileimage[12] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0073.jpg", 1); // Read the fileimage[13] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0074.jpg", 1); // Read the fileimage[14] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0075.jpg", 1); // Read the fileimage[15] = imread("C:/Users/Chen/Desktop/OpenCV_test/OpenCV_test/OpenCV_test/memorial0076.jpg", 1); // Read the fileMat output;Scalar     mean;Scalar     stddev;float mean_gray[16];  //灰度均值//获取灰度均值for (i = 0; i < 16; i++){meanStdDev(image[i], mean, stddev);  //分别获取图的均值与方差，在本算法中仅使用了均值mean_gray[i] = (mean[0] + mean[1] + mean[2])/3;cout << "<----mean_pxl------>" << mean_gray[i] << "<-------stddev_pxl------->" << stddev[0] << endl;}//通过灰度均值获取权值float weight[16];float sum=0;float sum2=0;float yu[16];for (i = 0; i < 16; i++){yu[i] = mean_gray[i] - 128;if (yu[i] < 0){yu[i] = - yu[i];}sum += yu[i]; }for (i = 0; i < 16; i++){sum2 += sum / yu[i];}for (i = 0; i < 16; i++){weight[i] =( sum / yu[i] )/sum2;cout << "权值" << weight[i];}output =  image[0] * weight[0]+ image[1] * weight[1]+ image[2] * weight[2]+ image[3] * weight[3]+ image[4] * weight[4]+ image[5] * weight[5]+ image[6] * weight[6]+ image[7] * weight[7]+ image[8] * weight[8]+ image[9] * weight[9]+ image[10] * weight[10]+ image[11] * weight[11]+17;/*output = output + image[12] * weight[12] + image[13] * weight[13] + image[14] * weight[14] + image[15] * weight[15]+7;*/imshow("结果", output);waitKey(0);   // Wait for a keystroke in the windowreturn 0;
}

在写完了这一堆之后，我突然想到了一些事情，除了传统的RGB图像采用的R,G,B三个颜色通道来表示图像，这对于其色彩有着较好的表示，但是，对于曝光处理来说并不是最优解，如果应用HSV图像中的S（饱和度）和V（明度）来进行该算法的实现或许效果会更加优秀，这个坑等过几天有些事情（指调车）处理完了再填吧。哦对，还有一条走过的弯路，在最开始的时候，我曾挨个像素的进行融合，结果效果并不佳，我思考了一下，感觉应该是因为仅针对单个像素处理，无法顾及到整个图像的效果。