NSS with HVS - based image quality evaluation method
|Course||Circuits and Systems|
|Keywords||Image Quality Assessment Natural image statistics (NSS) Human visual system (HVS) Structural similarity Digital Watermarking No reference evaluation method|
Over the last decade, in the promotion of digital TV and consumer electronics applications, digital image and video processing technology has gained rapid development. The digital image quality assessment is one of the basic techniques of image video processing, algorithm analysis and comparison of system performance optimization, and communication image quality monitoring plays an important role. Natural image statistics (Natural Scene Statistics, NSS) model of the human visual system (Human Visual System, HVS) model, describe the essential attribute of the image quality observed in the guest and the host. As the theoretical basis of the papers to the NSS and HVS analysis of the law of natural image distortion signal characteristics, studies in various application conditions of the reference images and the reference image of the full reference image to the computer as a means accurate, fast, effective method to evaluate the quality of the digital image. The main research content and results are summarized as follows: 1) based on the visual perception of the image quality comprehensive. Weighted method based on HVS front-end features to improve the structural similarity measure (to SSIM). Based on visual masking effect, calculated distortion errors with the edge of the texture of visual sensitivity, get error visibility diagram images and content visibility map, a combination of both to get a weighted graph representing the image of local importance, weighted comprehensive SSIM as the overall objective quality . This method makes up the the SSIM ignored the HVS 低层 characteristics insufficient to improve the SSIM performance was evaluated for the quality of the images of the different type of distortion. 2) based on masking effect and distortion area by SSIM. The method according to HVS the distortion regional sensitivity different, the distortion of the edge image into a texture region, expansion region, and smooth area, calculate the regional average SSIM value, and the expansion region as best reflects the typical area of ??the image quality. Proposed SSIM distortion concepts and modeling the masking effect of the use of the smooth region SSIM compensation expansion area the SSIM, unified local distortion of the image with the global distortion of the image quality evaluation. The method combines zoning and masking effects, improved SSIM various distortion types of image quality evaluation. 3) The proposed evaluation method based on the distribution of the energy spectrum distortion reference. The measure of image quality, to the image of the natural extent of the energy spectrum distribution of the important natural image statistical rule (NSS), the image distortion and the degree of variation of the distribution and distortion proportional to the intensity change. Contourlet transform domain, the use of low-scale sub-band energy of distortion robust high-scale sub-band energy forecast, the calculated subband energy distortion degree and can effectively predict JPEG2000 compression, Gaussian Blur, Gaussian white noise and other types of distortion image quality. Also proposed is a JPEG compression distortion metric method, the ratio of the energy in the image block boundary within the energy block, effectively predict the blockiness and JPEG compression strength, reflecting the characteristics of the JPEG compression to change the energy spatial distribution. 4) semi-reference quality evaluation of image quality key feature extraction and transmission method study. The analysis and comparison of the performance as the evaluation of the quality of the key features of a variety of sub-band and subband statistics rms contrast is one of the most efficient and simple to reflect the image quality of the image features. Digital watermarking is the major mode of transmission of the image quality characteristic, a new content authentication watermarking method, the method extracts a motion region in the image as the content key region, according to the importance of different authentication watermark for the content providing sub-levels of protection, and resistant to local tampering, MPEG compression and other attacks.