《学习OpenCV》练习题第五章第一题ab

这道题是载入一幅带有有趣纹理的图像并用不同的模板（窗口，核）大小做高斯模糊（高斯平滑），然后比较用5*5大小的窗口平滑图像两次和用11*11大小的窗口平滑图像一次是否接近相同。

先说下我的做法，a部分我将每个不同的窗口大小模糊化后的图像生成后，还计算了每个模糊化后的图像与原始图像间的MSE值与PSNR值。（参见：http://zh.wikipedia.org/wiki/%E5%B3%B0%E5%80%BC%E4%BF%A1%E5%99%AA%E6%AF%94）

b部分我计算了两次5*5窗口大小的高斯模糊后的图像与一次11*11窗口大小的高斯模糊图像之间的MSE与PSNR。

代码：

 #include <opencv/highgui.h>
 #include <opencv/cv.h>
 #include <opencv_libs.h>
 #include <math.h>

 /*
  *《学习OpenCV》第五章第一题
  * 完成时间：18:37 10/13 星期日 2013
  * 作者：[e-mail]
  */

 /*
  * function: calculate MSE & PSNR of two GrayScale(8-bit depth & one channel) images.
  * param: img1 -- the first image.
  * param: img2 -- the second image.
  * param: dMSE -- the MSE of two images(output)
  * param: dPSNR -- the PSNR of two images(output)
  * return: 0 -- success;  others -- failed.
  */
 int calculateGrayImgsPSNR(IplImage* img1, IplImage* img2, double& dMSE, double& dPSNR)
 {
     if( !img1 || !img2 ||
         img1->nChannels !=  ||
         img2->nChannels !=  ||
         img1->depth != img2->depth ||
         img1->width != img2->width ||
         img1->height != img2->height )
     {
         return -;
     }
     int width = img1->width;
     int height = img1->height;

     // calculate MSE of the two images
     double dSumOfSquares = ;
     for(int i = ; i < height; i++)
     {
         char* pdata1 = img1->imageData + i * img1->widthStep;
         char* pdata2 = img2->imageData + i *img2->widthStep;
         for(int j = ; j < width; j++ )
         {
             uchar value1 = *(pdata1 + j);
             uchar value2 = *(pdata2 + j);

             double square = pow( (double)(value1 - value2),  );
             dSumOfSquares += square;
         }
     }

     dMSE = dSumOfSquares / (width * height);

     // this is means the two images are strictly same.
     if(dMSE == )
     {
         dPSNR = -;
         return ;
     }
     int iDepth = img1->depth;
     int iMAX = pow( ., iDepth) - ;

     dPSNR =  * log10(iMAX / (sqrt(dMSE)));

     return ;
 }

 int main()
 {
     const char * FILE_PATH = "Fig0333(a)(test_pattern_blurring_orig).tif";

     IplImage* src = cvLoadImage(FILE_PATH, CV_LOAD_IMAGE_UNCHANGED);

     if(!src)
     {
         printf("Load image error.\n");
         return -;
     }

     // Get the source image's size
     CvSize srcSize = cvGetSize(src);

     // 3 * 3
     IplImage* dst_three_gaussian = cvCreateImage(srcSize, src->depth, src->nChannels);
     // 5 * 5
     IplImage* dst_five_gaussian = cvCreateImage(srcSize, src->depth, src->nChannels);
     // 9 * 9
     IplImage* dst_nine_gaussian = cvCreateImage(srcSize, src->depth, src->nChannels);
     // 11 * 11
     IplImage* dst_eleven_gaussian = cvCreateImage(srcSize, src->depth, src->nChannels);
     // twice 5 * 5
     IplImage* dst_twice_five_gaussian = cvCreateImage( srcSize, src->depth, src->nChannels );

     if( !dst_three_gaussian || !dst_five_gaussian ||
         !dst_nine_gaussian || !dst_eleven_gaussian ||
         !dst_twice_five_gaussian )
     {
         printf("Create image error.\n");
         return -;
     }

     cvSmooth(src, dst_three_gaussian, CV_GAUSSIAN, , );
     cvSmooth(src, dst_five_gaussian, CV_GAUSSIAN, , );
     cvSmooth(src, dst_nine_gaussian, CV_GAUSSIAN, , );
     cvSmooth(src, dst_eleven_gaussian, CV_GAUSSIAN, , );
     cvSmooth( dst_five_gaussian, dst_twice_five_gaussian, CV_GAUSSIAN, ,  );

     cvShowImage("src", src);
     cvShowImage("src - GAUSSIAN 3*3", dst_three_gaussian);
     cvShowImage("src - GAUSSIAN 5*5", dst_five_gaussian);
     cvShowImage("src - GAUSSIAN 9*9", dst_nine_gaussian);
     cvShowImage("src - GAUSSIAN 11*11", dst_eleven_gaussian);
     cvShowImage("src - GAUSSIAN 5*5 Twice", dst_twice_five_gaussian );

     // calculate the MSE and PSNR of the two images.
     double dMSE, dPSNR;
     // part a:
     calculateGrayImgsPSNR(src, dst_three_gaussian, dMSE, dPSNR);
     printf("source image & 3*3 GAUSSIAN: MSE: %f\tPSNR: %f\n", dMSE, dPSNR);
     calculateGrayImgsPSNR(src, dst_five_gaussian, dMSE, dPSNR);
     printf("source image & 5*5 GAUSSIAN: MSE: %f\tPSNR: %f\n", dMSE, dPSNR);
     calculateGrayImgsPSNR(src, dst_nine_gaussian, dMSE, dPSNR);
     printf("source image & 9*9 GAUSSIAN: MSE: %f\tPSNR: %f\n", dMSE, dPSNR);
     calculateGrayImgsPSNR(src, dst_eleven_gaussian, dMSE, dPSNR);
     printf("source image & 11*11 GAUSSIAN: MSE: %f\tPSNR: %f\n", dMSE, dPSNR);

     // part b
     puts("---------------------------\n");
     calculateGrayImgsPSNR(src, dst_eleven_gaussian, dMSE, dPSNR);
     printf("source image & eleven: MSE: %f\tPSNR: %f\n", dMSE, dPSNR);
     calculateGrayImgsPSNR(src, dst_twice_five_gaussian, dMSE, dPSNR);
     printf("source image & twice five: MSE: %f\tPSNR: %f\n", dMSE, dPSNR);
     calculateGrayImgsPSNR(dst_eleven_gaussian, dst_twice_five_gaussian, dMSE, dPSNR);
     printf("eleven & twice five: MSE: %f\tPSNR: %f\n", dMSE, dPSNR);

     cvWaitKey();
     cvReleaseImage(&src);
     cvReleaseImage(&dst_three_gaussian);
     cvReleaseImage(&dst_five_gaussian);
     cvReleaseImage(&dst_nine_gaussian);
     cvReleaseImage(&dst_eleven_gaussian);
     cvReleaseImage(&dst_twice_five_gaussian);
     cvDestroyAllWindows();

     return ;
 }

运行结果：

a部分：

3*3：

5*5：

9*9：

11*11：

同时各个不同的窗口大小模糊化后的图像与原始图像之间的MSE与PSNR：

从图中可以看出，当窗口大小越大时，MSE增大，PSNR减小。

b部分：

两幅图像的PSNR与MSE：

众所周知的是在图像压缩中典型的PSNR比值在30-40dB之间，而我们这两幅平滑之后的图像PSNR为31.976534，所以这两幅图像是比较接近的。