Image Quality Assessment Based On Contourlet And Gradient Structural Similarity

Image quality assessment has been separated from the image processing to become anindependent field of study. The ultimate recipient of the image information is a person, so thesubjective method is reliable. But it requires repeated experiments, time-consuming, difficultto be applied to practical applications. Objective image quality assessment method hasbecome a hot research, traditional objective evaluation method is simple, but because itdoesn’t consider the human visual perception and image characteristics, often is out of touchwith people’s subjective evaluation. Z.Wang combined the structural distortion theory andproposed structural similarity (SSIM, structural similarity) algorithm.The algorithm issuperior, but there is also drawback.SSIM algorithm is analysed in this paper. The superiority of SSIM has been verifiedthrough experimental result.The SSIM also pointed out that there are some errors in theevaluation with different types of distortion image for the same image. The main problemexists in severely blurred image. The texture characteristics of blurred image are destroyed,the structure(s) component of SSIM methodand does not accurately reflect the change. Thegradient of the image can be well reflected in the edge and the texture transformation. Followthe existing gradirent structure similarity algorithm; consider Contourlet has goodcharacteristics similar to that of the human visual system (HVS). Contourlet has excellentcharacteristics of the local, multi-scale, multi-direction. This paper proposes an image qualityassessment algorithm based on Contourlet and gradient structural similarity (CGSSIM).In this method, the gradient structure similarity (GSSIM) algorithm is used to eachband-pass directional subband of Contourlet decomposition. Gradient structure similarities ofall directional subbands are weighted, and get the Contourlet transform domain of the wholeimage gradient structure similarity (CGSSIM). Weights experimentally determine thesensitivity of the human eye with different band-pass directional subband. By the experiment,the CGSSIM algorithm is proved that can get better results with an image of the Gaussian blur and the image of white noise distortion cross evaluation. The experimental results also showthat the CGSSIM algorithm is more suitable for the human visual system characteristics thanthe PSNR and SSIM algorithm, and it is closer with the subjective perception, the accuracy ofevaluation and consistency better. Especially Gaussian blur image evaluation, its accuracy hasbeen improved to some extent.