Based On Structural Similarity And Sparse Repres-Entation Research On FR_IQA Algorithm

With the rapid development of Internet technology,digital images are widely used in all walks of life.However,due to the image acquisition,transmission,compression,etc.,it is very easy to produce distortion,resulting in a decline in image quality.Therefore,choosing an appropriate image quality evaluation algorithm to evaluate the distorted image has high research and application value.In this paper,three aspects of research work are carried out around the evaluation of full reference image quality.(1)An improved structural similarity DS_SSIM algorithm is proposed.In the SSIM algorithm,the image blocks are treated equally,and the pixel-by-pixel comparison is performed,which is inconsistent with the human visual system.this paper judges the sensitivity of human visual perception and introduces it into the SSIM model.The four algorithms of PSNR,SSIM,CSSIM,and DS_SSIM were tested on the LIVE database,and the five distortion types of JPEG,JPEG2000,Fast Fading,Gaussian Blur,and White Noise,and all the distortion maps of the entire LIVE image library were experimentally evaluated.The results It shows that the DS_SSIM algorithm shows better experimental performance.(2)An improved gradient structure similarity GDSM algorithm is proposed.Considering the problem of different quality of different local structures in the distorted image,the distortion pixel measure is added to the GSSIM model,and the proposed GDSM algorithm.The algorithm and the five algorithms of PSNR,SSIM,GSSIM,FSIM and DS_SSIM were respectively tested and evaluated on the three databases of LIVE,TID2008 and TID2013.The experimental results show that the GDSM algorithm overcomes the GSSIM algorithm to directly take each image sub-block The average value of the similarity of the gradient structure is taken as the shortcoming of the overall image quality evaluation value,Compared with the DS_SSIM algorithm,the fitted scatter distribution is more aggregated,indicating that the GSDM algorithm is more consistent with the HVS system.(3)An improved adaptive sub-dictionary PC-OASD algorithm is proposed.In the algorithm based on sparse representation,only the sparse features are extracted to capture the structural changes between the reference image and the distorted image,but due to the weak sparse features of the image Distortion is not sensitive,and the perceived image quality is generally affected by weak distortion.Therefore,while extracting sparse features,this paper adds three auxiliary features:gradient,phase consistency,and brightness,which are described together by weighted sparse features.Finally,in three databases LIVE,TID2008 and TID2013,the algorithm was compared with PSNR,SSIM,VIF,MAD,FSIM,GMSD,DS_SSIM,GSSIM,and QASD algorithms.The correlation coefficients were compared and verified.The results showed that:The QASD algorithm is superior to the rest of the algorithms,and it performs well on all databases.