java opencv dct 压缩_VS2010实现opencv基于DCT的图像压缩

Refer from http://blog.csdn.net/fjssharpsword/article/details/7765908

1.基于DCT的图像压缩方法是将一幅图像分割成矩形像素块，再分别对每一独立的像素块DCT变换、量化、编码和传输。

2.打开VS2010，新建Visual C++下Win32控制台应用程序demo，主程序如下：

// demo.cpp : 定义控制台应用程序的入口点。

//

#include "stdafx.h"

#include "highgui.h"

#include

#include

#include "cxcore.h"

#define cvCvtPlaneToPix cvMerge

double PSNR_B = 0;

double PSNR_G = 0;

double PSNR_R = 0;

double PSNR;

int _tmain(int argc, _TCHAR* argv[])

{

const char* imagename = "D:/demo/demo.jpg";

IplImage *src;

CvScalar SrcPixel;

CvScalar DstPixel;

double sumB = 0;

double sumG = 0;

double sumR = 0;

double mseB;

double mseG;

double mseR;

src= cvLoadImage( imagename,1 ) ;

if( !src )

{

printf("can't open the image...\n");

return -1;

}

// YUV颜色空间

IplImage* YUVImage = cvCreateImage(cvSize(src->width,src->height), src->depth, 3);

IplImage* dst = cvCreateImage(cvSize(src->width,src->height), src->depth, 3);

// YUV颜色空间各通道

IplImage* Y = cvCreateImage(cvSize(src->width,src->height), IPL_DEPTH_8U, 1);

IplImage* U = cvCreateImage(cvSize(src->width,src->height), IPL_DEPTH_8U, 1);

IplImage* V = cvCreateImage(cvSize(src->width,src->height), IPL_DEPTH_8U, 1);

//cvNamedWindow( "Origin Image", CV_WINDOW_AUTOSIZE );

cvCvtColor(src, YUVImage, CV_BGR2YUV); //BGR→YUV

cvSplit( YUVImage, Y, U, V, NULL);//分割通道

CvMat* MatY = cvCreateMat(Y->height,Y->width,CV_64FC1);

CvMat* MatU = cvCreateMat(V->height,U->width,CV_64FC1);

CvMat* MatV = cvCreateMat(V->height,V->width,CV_64FC1);

CvMat* DCTY = cvCreateMat(Y->height, Y->width,CV_64FC1);

CvMat* DCTU = cvCreateMat(U->height, U->width,CV_64FC1);

CvMat* DCTV = cvCreateMat(V->height, V->width,CV_64FC1);

cvScale( Y, MatY );

cvScale( U, MatU );

cvScale( V, MatV );

cvDCT(MatY, DCTY, CV_DXT_FORWARD); //余弦变换

cvDCT(MatU, DCTU, CV_DXT_FORWARD); //余弦变换

cvDCT(MatV, DCTV, CV_DXT_FORWARD); //余弦变换

//Y 通道压缩

for(int i = 0; i < Y->height; i++)

{

for(int j = 0; j < Y->width; j++)

{

double  element = CV_MAT_ELEM( *DCTY, double, i, j );

if ( abs(element) < 10 )

CV_MAT_ELEM( *DCTY, double, i, j ) = 0;

}

}

// U 通道压缩

for(int i = 0; i < U->height; i++)

{

for(int j = 0; j < U->width; j++)

{

double  element = CV_MAT_ELEM( *DCTU, double, i, j );

if (     abs(element) < 20 )

CV_MAT_ELEM( *DCTU, double, i, j ) = 0;

}

}

// V 通道压缩

for(int i = 0; i < V->height; i++)

{

for(int j = 0; j < V->width; j++)

{

double  element = CV_MAT_ELEM( *DCTV, double, i, j );

if (     abs(element) < 20 )

CV_MAT_ELEM( *DCTV, double, i, j ) = 0;

}

}

cvDCT(DCTY, MatY, CV_DXT_INVERSE); //余弦反变换

cvDCT(DCTU, MatU, CV_DXT_INVERSE);

cvDCT(DCTV, MatV, CV_DXT_INVERSE);

cvScale( MatY, Y );

cvScale( MatU, U );

cvScale( MatV, V );

cvMerge( Y, U, V, NULL, YUVImage );

cvCvtColor( YUVImage, dst, CV_YUV2BGR); //YUV→BGR

//  计算前后两幅图像的PSNR值

for(int i = 0; i < src->height; i++)

{

for(int j = 0; j < src->width; j++)

{

SrcPixel = cvGet2D( src, i, j );

DstPixel = cvGet2D( dst, i, j );

sumB += ( SrcPixel.val[0] - DstPixel.val[0] ) * ( SrcPixel.val[0] - DstPixel.val[0] );

sumG += ( SrcPixel.val[1] - DstPixel.val[1] ) * ( SrcPixel.val[1] - DstPixel.val[1] );

sumR += ( SrcPixel.val[2] - DstPixel.val[2] ) * ( SrcPixel.val[2] - DstPixel.val[2] );

}

}

mseB = sumB / ((src->width) * (src->height)); //计算均方差

mseG = sumG / ((src->width) * (src->height));

mseR = sumR / ((src->width) * (src->height));

PSNR_B = 10.0 * ( log10( 255.0 * 255.0 / mseB ) );

PSNR_G = 10.0 * ( log10( 255.0 * 255.0 / mseG ) );

PSNR_R = 10.0 * ( log10( 255.0 * 255.0 / mseR ) );

PSNR=(PSNR_B + PSNR_G + PSNR_R) / 3;

cvShowImage( "YImage", Y );

cvShowImage( "UImage", U );

cvShowImage( "VImage", V );

cvShowImage( "DstImage", dst );

cvSaveImage( "D:/demo/dstdemo.jpg", dst);

while( 1 )

{

if( cvWaitKey(0) == 27 ) break;

}

cvDestroyWindow("YImage");

cvDestroyWindow("UImage");

cvDestroyWindow("VImage");

cvDestroyWindow("DstImage");

cvReleaseImage(&Y);

cvReleaseImage(&U);

cvReleaseImage(&V);

cvReleaseImage(&src);

cvReleaseImage(&dst);

cvReleaseImage(&YUVImage);

return 0;

}

3.参考用，jpg图片越压缩越大，代码有待优化来达到压缩效果！