Variation Image Denoising And Quality Assessment Based On Human Visual System

An image will be affected by noise in the process of the image generation andtransmission, this will affect the image visual effect. If the image’s basic characteristics canbe guaranteed completely, and further improve about the image visual effect will providebetter image data for subsequent calculation and digital image processing, it can bring greatconvenience about the use of the image.The main content of this paper is design image denoise and quality evaluation algorithmbased on the human visual characteristics, the main work and innovation as follows:Firstly, the human visual characteristics-fovea, contrast sensitivity, masking effect isdescribed. By this visual perception characteristics, the process of human visual perceptioncan be effectively simulated, furthermore, image denoise algorithms can be designed, this canprovide evidence for the subsequent image processing.Secondly, this paper describes the image quality assessment algorithms-PSNR (peaksignal-to-noise ratio),SSIM (structural similarity). Because the previous SSIM assumptionsdoes not be hold for small patches, meanwhile, when the variance of noise is significantcompared with the image patch, SSIM is in favor of the noise patterns rather than the imagestructures. In this paper, the SSIM evaluation algorithm is optimized and two-stage method isproposed. The improved algorithm based on SSIM reflects more similar characteristics to thehuman visual, it shows the superiority of the algorithm.Finally, in connection with the ladder phenomenon that existing in variational model,visual observation can be effected. This paper improve the variational model, a differentregularization of variational method is presented, so that the noise reduction effect is moreconsistent with human visual characteristics.In addition,because the PM diffusion model blurred the image edge, the diffusionfunction is improved and the gradient threshold is set，so that the edge and texture of theimage can be retained very well. The strength coefficient is added in the diffusion model,bothit can enhance the image details and can also have better visual effects in low contrast image.Using the optimized SSIM evaluation algorithm to evaluation the improved variationaldenoise model and compared with the ROF model, the results show that the denoise algorithmproposed in this paper have made a better denoise effect both at the contour edge and texture details.