Research On Methods Of Objective Image Quality Assessment

With the development of information technology,the application of digital images in daily production and life is everywhere.Various types of distortion may be introduced in various aspects of image communication,resulting in loss of image quality and interference with subjective perception.In order to meet the image quality requirements of different applications,it is necessary to establish an efficient objective quality assessment mechanism to simulate the subjective perception process of human visual system(HVS).Most of the existing objective evaluation methods only deal with the luminance component of the color image,ignoring the influence of the color information on the image quality perception.The introduction of color information in the objective evaluation method can obtain better evaluation performance.Two color image quality assessment methods are proposed in this thesis:(1)A full reference color image quality assessment method based on two scale feature similarity weighting.Features extracted on a single scale can only reflect certain characteristics of HVS and have very limited ability to characterize image quality.This thesis designs a two scale feature similarity weighted quality index(TSFSWQI),which first transforms the color space and separates the luminance and color components.Then,based on the pixel scale,multiple features are extracted.The similarity metric on the pixel scale is obtained by the operations of fusion,weighting and averaging;then the statistical features are extracted based on the image scale,the feature vector is constructed,and the Euclidean distance of the feature vector is calculated as the similarity metric on the image scale;finally,the similarity metrics of the two scale are weighted to obtain the TSFSWQI score.Experiments show that TSFSWQI has high consistency with subjective assessment and robustness when it evaluates images across different databases and distortion types.(2)A no reference color image quality assessment method based on local binary pattern and local entropy.In this thesis,we improve the performance of the spatial-spectral entropy based quality index(SSEQ),and the local binary pattern features of YIQ components are introduced respectively to design a local binary pattern and spatial-spectral entropy based quality index(LBPSSEQ).Experiments show that the LBPfeatures can effectively reflect the image quality changes caused by distortion,which makes the performance of LBPSSEQ significantly better than SSEQ,and has a relatively good comprehensive performance compared with similar methods.