参考:http://blog.csdn.net/trent1985/article/details/50904173
根据国外一篇大牛的文章:No-Reference Perceptual Quality Assessment of JPEG Compressed Images
在无参考图像的质量评价中,图像的清晰度是衡量图像质量优劣的重要指标,它能够较好的与人的主观感受相对应,图像的清晰度不高表现出图像的模糊。本文针对无参考图像质量评价应用,对目前几种较为常用的、具有代表性清晰度算法进行讨论分析,为实际应用中选择清晰度算法提供依据。
对于JPEG图像,根据大牛的文章,写成的MATLAB代码如下:
function score = jpeg_quality_score(img)
%========================================================================
%
%Copyright (c) 2002 The University of Texas at Austin
%All Rights Reserved.
%
%This program is free software; you can redistribute it and/or modify
%it under the terms of the GNU General Public License as published by
%the Free Software Foundation; either version 2 of the License, or
%(at your option) any later version.
%
%This program is distributed in the hope that it will be useful,
%but WITHOUT ANY WARRANTY; without even the implied warranty of
%MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
%GNU General Public License for more details.
%
%The GNU Public License is available in the file LICENSE, or you
%can write to the Free Software Foundation, Inc., 59 Temple Place -
%Suite 330, Boston, MA 02111-1307, USA, or you can find it on the
%World Wide Web at http://www.fsf.org.
%
%Author : Zhou Wang
%Version : 1.0
%
%The authors are with the Laboratory for Image and Video Engineering
%(LIVE), Department of Electrical and Computer Engineering, The
%University of Texas at Austin, Austin, TX.
%
%Kindly report any suggestions or corrections to zhouwang@ieee.org
%
%========================================================================
%
%This is an implementation of the algorithm for calculating the quality
%score of JPEG compressed images proposed by Zhou Wang, Hamid R. Sheikh
%and Alan C. Bovik. Please refer to the paper: Zhou Wang, Hamid R. Sheikh
%and Alan C. Bovik, "No-Reference Perceptual Quality Assessment of JPEG
%Compressed Images," submitted to IEEE International Conference on Image
%Processing, Sept. 2002.
%
%You can change this program as you like and use it anywhere, but please
%refer to its original source (cite our paper and our web page at
%http://anchovy.ece.utexas.edu/~zwang/research/nr_jpeg_quality/index.html).
%
%Input : A test 8bits/pixel grayscale image loaded in a 2-D array
%Output: A quality score of the image. The score typically has a value
% between 1 and 10 (10 represents the best quality, 1 the worst).
%
%Usage:
%
%1. Load the image, for example
%
% image = imread('testimage.jpg');
%
%2. Call this function to calculate the quality score:
%
% Quality_Score = jpeg_quality_score(image)
%
%========================================================================
% img=imread('/Users/anitafang/Desktop/testimg/test0.jpg');
if (nargin > 1)
score = -1;
return;
end
% [M,N] = size(img);
% if (M < 16 | N < 16)
% score = -2;
% return;
% end
x1= rgb2gray(img);
x = double(x1);
[M,N] = size(x);
% Feature Extraction:
% 1. horizontal features
d_h = x(:, 2:N) - x(:, 1:(N-1));
% [m, n]=size(d_h);
% disp(d_h);
% fprintf('img width %d,and height %d,length %d, dims %d\n',M,N,length(img),ndims(img));
% fprintf('d_h width %d,and height %d, length %d, dims %d\n',m,n,length(d_h),ndims(d_h));
B_h = mean2(abs(d_h(:, 8:8:8*(floor(N/8)-1))));
A_h = (8*mean2(abs(d_h)) - B_h)/7;
sig_h = sign(d_h);
left_sig = sig_h(:, 1:(N-2));
right_sig = sig_h(:, 2:(N-1));
Z_h = mean2((left_sig.*right_sig)<0);
% fprintf('B_h:%f,A_h:%f,Z_h:%f,\n',B_h,A_h,Z_h);
% 2. vertical features
d_v = x(2:M, :) - x(1:(M-1), :);
B_v = mean2(abs(d_v(8:8:8*(floor(M/8)-1), :)));
A_v = (8*mean2(abs(d_v)) - B_v)/7;
sig_v = sign(d_v);
up_sig = sig_v(1:(M-2), :);
down_sig = sig_v(2:(M-1), :);
Z_v = mean2((up_sig.*down_sig)<0);
% 3. combined features
B = (B_h + B_v)/2;
A = (A_h + A_v)/2;
Z = (Z_h + Z_v)/2;
% Quality Prediction
alpha = -245.8909;
beta = 261.9373;
gamma1 = -239.8886;
gamma2 = 160.1664;
gamma3 = 64.2859;
score = alpha + beta*(B.^(gamma1/10000))*(A.^(gamma2/10000))*(Z.^(gamma3/10000));
调用的main函数:
function main() allNum = 0; blurNum = 0; threshold = 7.7; for k = 0:2000 try image = imread(['/Users/anitafang/Desktop/testimg/','test',num2str(k),'.jpg']); %image = imread(['/Users/user/Desktop/test/ye_blur/',num2str(k),'.jpg']); quality_score = jpeg_quality_score(image); fprintf('Quality_Score:%d.jpg %f\n',k,quality_score); allNum = allNum + 1; if quality_score < threshold blurNum = blurNum + 1; savePath = ['/Users/anitafang/Desktop/blur/',num2str(k),'.jpg']; imwrite(image,savePath); end % break catch err %throw(err); end end % fprintf('relevance:%f\n',blurNum/allNum)
为了集成方便,把它变成c++代码,调研opencv库:
// // jpegquality.hpp // SDM_Train // // Created by anitafang on 2017/6/27. // Copyright © Anita,fang. All rights reserved. // #ifndef jpegquality_hpp #define jpegquality_hpp #include <stdio.h> #include <vector> #include <iostream> #include <time.h> #include <fstream> #include <math.h> #include <cmath> #include "opencv2/opencv.hpp" #include "opencv2/core/core.hpp" #include "opencv2/highgui/highgui.hpp" #include "opencv2/objdetect/objdetect.hpp" #endif /* jpegquality_hpp */ using namespace std; using namespace cv; class JPEGQ{ public: JPEGQ(); void h_feature(Mat x,int M,int N); void v_feature(Mat x,int M,int N); void combine_feature(); double qual_predict(); private: double B_h,A_h,Z_h; double B_v,A_v,Z_v; double B,A,Z; };
cpp代码是:
// // jpegquality.cpp // SDM_Train // // Created by anitafang on 2017/6/27. // Copyright © 2017年 antia.fang All rights reserved. // #include "jpegquality.hpp" using namespace std; using namespace cv; JPEGQ::JPEGQ(){ B_h = 0; A_h = 0; Z_h = 0; B_v = 0; A_v = 0; Z_v = 0; B = 0; A = 0; Z = 0; } //% 1. horizontal features void JPEGQ::h_feature(Mat x,int M,int N){ Mat d_h(M,N-1,CV_64F, Scalar(0,0,255)); for(int i=0;i<M;i++){ for(int j=0;j<N-1;j++){ d_h.at<double_t>(i,j)=x.at<double_t>(i,j+1)-x.at<double_t>(i,j); } } // cout << "d_h = " << d_h << "\n"; int DEL= floor(N/8); Mat d1_h(M,DEL,CV_64F, Scalar(0,0,255)); for (int i=0;i<M;i++){ for(int j=0;j<DEL;j++){ d1_h.at<double_t>(i,j)= abs(d_h.at<double_t>(i,8*j+7)); } } //求矩阵像素的的平均值 B_h = mean(d1_h)[0]; A_h = (8*mean(abs(d_h))[0] - B_h)/7; Mat sig_h(M,N-2,CV_64F, Scalar(0,0,255)); for(int i=0;i<M;i++){ for(int j=0;j<N-1;j++){ if(d_h.at<double_t>(i,j) < 0){ sig_h.at<double_t>(i,j) = -1; } else if(d_h.at<double_t>(i,j) > 0){ sig_h.at<double_t>(i,j) = 1; } else { sig_h.at<double_t>(i,j) = 0; } } } Mat left_sig(M,N-2,CV_64F,Scalar(0,0,255)); Mat right_sig(M,N-2,CV_64F,Scalar(0,0,255)); for(int i=0;i<M;i++){ for(int j=0;j<N-2;j++){ left_sig.at<double_t>(i,j)=sig_h.at<double_t>(i,j); right_sig.at<double_t>(i,j)=sig_h.at<double_t>(i,j+1); } } //double Z_h = mean2((left_sig.*right_sig)<0); Mat multi_sig(M,N-2,CV_64F, Scalar(0,0,255)); for(int i=0;i<M;i++){ for(int j=0;j<N-2;j++){ double temp1=left_sig.at<double_t>(i,j)* right_sig.at<double_t>(i,j); if(temp1<0){ multi_sig.at<double_t>(i,j)= 1; } else { multi_sig.at<double_t>(i,j)= 0; } } } Z_h =mean(multi_sig)[0]; // cout <<"B_h: "<< B_h<<"A_h: "<< A_h<<"Z_h: "<< Z_h << endl; } // % 2. vertical features void JPEGQ::v_feature(Mat x,int M,int N){ Mat d_v(M-1,N,CV_64F, Scalar(0,0,255)); for(int i=0;i<M-1;i++){ for(int j=0;j<N;j++){ d_v.at<double_t>(i,j)=x.at<double_t>(i+1,j)-x.at<double_t>(i,j); } } int DELV= floor(M/8); Mat d1_v(DELV,N,CV_64F, Scalar(0,0,255)); for (int i=0;i<DELV;i++){ for(int j=0;j<N;j++){ d1_v.at<double_t>(i,j)= abs(d_v.at<double_t>(8*i+7,j)); } } //求矩阵像素的的平均值 B_v=mean(d1_v)[0]; A_v = (8*mean(abs(d_v))[0] - B_v)/7; Mat sig_v(M-1,N,CV_64F, Scalar(0,0,255)); for(int i=0;i<M-1;i++){ for(int j=0;j<N;j++){ if(d_v.at<double_t>(i,j)<0) { sig_v.at<double_t>(i,j)=-1; } else if(d_v.at<double_t>(i,j) >0) { sig_v.at<double_t>(i,j) = 1; } else { sig_v.at<double_t>(i,j) = 0;} } } Mat up_sig(M-2,N,CV_64F, Scalar(0,0,255)); Mat down_sig(M-2,N,CV_64F, Scalar(0,0,255)); for(int i=0;i<M-2;i++){ for(int j=0;j<N;j++){ up_sig.at<double_t>(i,j)=sig_v.at<double_t>(i,j); down_sig.at<double_t>(i,j)=sig_v.at<double_t>(i+1,j); } } //double Z_h = mean2((left_sig.*right_sig)<0); Mat vmulti_sig(M-2,N,CV_64F, Scalar(0,0,255)); for(int i=0;i<M-2;i++){ for(int j=0;j<N;j++){ double temp2=up_sig.at<double_t>(i,j)* down_sig.at<double_t>(i,j); if(temp2<0) { vmulti_sig.at<double_t>(i,j)= 1;} else { vmulti_sig.at<double_t>(i,j)= 0; } } } Z_v =mean(vmulti_sig)[0]; } //% 3. combined features void JPEGQ::combine_feature(){ B = (B_h + B_v)/2; A = (A_h + A_v)/2; Z = (Z_h + Z_v)/2; } //% Quality Prediction double JPEGQ::qual_predict(){ double alpha = -245.8909; double beta = 261.9373; double gamma1 = -239.8886; double gamma2 = 160.1664; double gamma3 = 64.2859; double score = alpha + beta*(pow(B,gamma1/10000)*pow(A,gamma2/10000)*pow(Z,gamma3/10000)); return score; }
调用的main代码:
// // main.cpp // jpg_quality // // Created by anitafang on 2017/6/28. // Copyright © 2017年 anitafang. All rights reserved. // #include <iostream> #include "jpegquality.hpp" using namespace std; using namespace cv; int main(int argc, const char * argv[]) { char filename[100]; double threshold = 7.7; int num =12; // int *pia = new int[num] (); // 每个元素初始化为0 for(int k=0;k<num;k++){ sprintf(filename,"/Users/anitafang/Desktop/VIP/xcode-demo/test-img/testimg/test%d.jpg",k); Mat x1=imread(filename,IMREAD_GRAYSCALE); int M=x1.rows; int N=x1.cols; Mat x; x1.convertTo(x, CV_64F);//转换成浮点运算 JPEGQ *jpegq = new JPEGQ(); jpegq->h_feature(x,M, N); jpegq->v_feature(x,M, N); jpegq->combine_feature(); double score= jpegq->qual_predict(); if(score<threshold){ cout<<"this is a blur image"<<endl; } cout<<"the image :"<<k<<" "<<" score is :"<<score<<endl; } return 0; }
可以看到阈值的设置7.7是个经验值。