An Image Quality Assessment Metric Based On Structure Similarity Of Extended Gradients

In recent years,China's communication technology has developed rapidly,and the amount of information that needs to be transmitted has also increased.For this reason,the requirements for information have also been continuously improved.Among many information carriers,images are increasingly important in the field of information interaction by virtue of their unique information-rich attributes.In the interactive part of image information,the image quality has been greatly valued.Therefore,for the application of image and video information,it is undoubtedly very significant to study the image quality evaluation technology.At present,there are two main categories of evaluating image quality: subjective and objective.For the former,it is the final evaluation based on the subjective feelings of people.This method has high accuracy and reliability,but it requires a lot of manpower and time,which makes the application scope not wide.For the latter,the mathematical model and related computer technology are used to scientifically evaluate the distortion maps.In this thesis,the classical objective evaluation algorithm is deeply analyzed.It is found that one of the key factors used by the gradient factor is the traditional gradient template.However,since only two directions are used,only horizontal and vertical gradients can be detected.At the edge of the direction,the detected edge direction is limited.For this reason,the extended gradient operator instead of the traditional gradient operator was used to be proposed to evaluate the image quality.The contents of this research are as follows:Firstly,the full reference image quality evaluation methods based on structural similarity(SSIM algorithm)and the gradient-based structural similarity full reference image quality evaluation method(GSSIM algorithm)are analyzed.By means of simulation and fitting performance analysis of these two algorithms,it is concluded that the latter algorithm has higher accuracy in the LIVE2 image database.Secondly,a structural similarity algorithm(E-GSSIM)based on extended gradient operator is proposed.Taking the GSSIM algorithm as the object,the limitations of the traditional Sobel operator are analyzed.Combined with the multi-directionality of the image edge,the direction of the traditional edge gradient operator is extended to 8 degrees,using 0°,22.5°,45°.,67.5°,90°,112.5°,135°,157.5° are calculated in eight directions,and then the extended Sobel operator is introduced to replace the past operator,and an extended quality evaluation algorithm,namely E-,is proposed.The GSSIM algorithm is then based on the database image of LIVE2 and so on.It is quantitatively and qualitatively analyzed,and the EGSSIM algorithm is obtained,which is better than the ordinary GSSIM algorithm.Thirdly,a similarity evaluation algorithm(MSPSIM)based on extended gradient operator under superpixel is proposed.The SPSIM algorithm is deeply analyzed and is based on the luminance factor,the chrominance factor and the gradient factor under the superpixel.The operator used by SPSIM in calculating the gradient factor is still the traditional gradient operator.For this reason,we extend the gradient factor in 8 directions under the condition of retaining the luminance factor and the chrominance factor,and propose corresponding similarity evaluation.The algorithm,which is in the corresponding four graphics databases such as LIVE,and the SPSIM algorithm performance has been improved in these databases,which also means that the overall performance of the algorithm is better.