The Reseach Of Similarity Based Image Quality Assessment Algorithm

With the development of science,technology and the popularization of the Internet,digital image technology has been widely used.The process of digital image acquisition,processed,coding,storage,transmission and reconstruction may cause loss of image quality.The loss of image quality affects the image expression and the accuracy of information which image is supposed to express,so the image quality assessment is important.When image process method continuously improves,high dynamic range image is widely accepted.Image assessment method needs constantly improvement so that it can be applied to new needs.Excellent digital image assessment method can not only assess the quality of the image,but also provide a good direction on improving image technology.This paper focuses on the theme of the full-reference image quality assessment that the corresponding reference image of the distorted image is available.Based on existing studies and research on the evaluation of image quality,this paper proposed our own image quality algorithms.First of all,based on the existing pixel level of weighted strategy,a second-level weighting method is proposed,which improved existing image quality assessment algorithms.Existing weighted strategy is concentrated on pixel level,human visual system tend to be interested in a particular area rather than pixels.Area-weighted algorithm is described as:local pixel level quality value is weighted by pixel weight in a regions,the quality of regional value is weight by region weight.Secondly,human eye is a very good information extractor.When there is distortion in image,the content information extracted by human visual system changes.In this paper,based on the information extraction characteristic of human eye,information content is modeled based on information theory.With image gradient and chromaticity characteristic,an information content similarity algorithm is proposed.Third,this paper modified gradient similarity and fuses it with naturalness to evaluate tone-mapped image.The test result on image quality assessment public database shows that all of the proposed algorithms achieve good performance.