Visual Perception Based Image And Video Quality Assessment

The aim of the image and video quality assessment is to find a computational model that can predict the perceptual visual quality automatically. An image is a vision entity that can describe the objective world in different forms by variable observation system. Compared with texts and graphics, the image conveys the information more vividly and affluently. Since images are subject to distortions during acquisition, compression, transmission, processing and reproduction, which will bring difficulties to the subsequent process such as image and video processing, analysis and understanding, It is not conducive to the correct understanding of the objective world. Thus, it is necessary to design or optimize an image and video processing method or system by measuring the degradation of the image and video, so that we can get the perfect visual quality with the least cost.In this dissertation, objecting to essential problems of the image and video quality assessment and exploring the relation between human visual system and digital information, objective assessment methods are built to evaluate some distortions by uncertain factors in process of the image and video, which are based on the image sparse representation and feature modeling for measuring visual quality of distorted image and the ability to provide information of the image, The main contributions of this dissertation can be summarized as follows:(1) According to the characteristic of the physiology and psychology of human visual system and combining error visibility models based on bottom-up approach, the variation of visual quality is measured by emplying the difference structure and geometrical similarity between reference images and distorted images from the basic structure and geometrical characteristic. A full reference image quality assessment method based on human visual system is proposed. Experimental results illustrate that the proposed has a good consistency with the subjective assessment of human beings, and it can be used to describe the visual perception of the image effectively.(2) According to feature extraction and sparse representation of the image and mimicing mutil-channel characteristic of human visual system, multiscale geometric analysis is employed to extract geometrical characteristics of the image. Then these geometrical characteristics are perceptually filtered based on the contrast sensitivity and masking of human visual system. The degradation of perceptual quality can be captured by these filtered features. So a framework of reduced reference image quality assessment based on multiscale geometric analysis is proposed, thorough empirical studies are carried out upon the LIVE database against subjective mean opinion score and demonstrate that the proposed framework has good consistency with subjective perception values and the objective assessment results can well reflect the visual quality of images.(3) On the basis of natural image statistic characteristics and making use of computational model to construct a generalized distribution of image features. Contourlet transform is introduced to decompose images and produce coefficients at difference subbands, relationship of whose coefficients are reflected by the mutual information among scale, direction and adjacent. By analyzing the variance between different scales of the image model and combing the diversification of the model , the variation degree of distortion is measured, and then a nonlinear mapping of the diversification of the features is used as the quality of the distorted image. A no reference image quality metric based on contourlet domain is proposed. This algorithm is demonstrated to be fit for various types of distortion, and is well consistent with the subjective perceptive results.(4) According to the features of different types of artifacts and the relation between visual perception and degradation process, we designed a new metric derived from the acuity and sensitivity of human eyes to image quality, analyzing the image global structure and local structure, to determine the degree of visual distortion which is based on the structure similarity of images. The experimental results have good correlation with visual perception. This method is extended to the video quality assessment based on the optimal distortion model. It also gives good performance.(5) The strengths and weaknesses of the VQEG video database have been well analyzed. In order to compensate the limitations of the database and catch up with the development of the current streaming media, a new video database is constructed based on the H.264 standard of the advanced video compression. Great deals of videos with rich content are compressed in different quantification scales and bit rates. A number of non-expert persons give the subjective scores for these videos on an automatic platform by subjective sorting methods. The new database offers a good beginning for further video quality assessment research.The research results above are scientific problems abstracted from practical application, referred to the basic theory of image and video processing. They are studied from new angles and based on new conditions and application, which are forward-looking and full of challenges and have extremely important theoretical significance and application value. The dissertation have some breakthrough in theory, some innovation in technology and some reference in methodology, it opens up a new way for image and video quality assessment.