Research On Image Quality Assessment Algorithm Based On Visual Information Fidelity

In recent years, the using of image is more and more widely in our life, such as surveillance video, 3D movie. With the growing demands in image applications, the users put forward much higher requirements for the quality of image, it pushes the development of the image quality assessment, whose goal is to design a mathematic model which can replace the human eye to accurately measure the image quality degradation. It has a very important significance on image processing. Base on the availability of reference images, the objective methods can be divided into full reference methods(FR), reduced reference methods(RF) and no reference methods(NF). In this paper, we focus on the FR image quality assessment methods. According to the perception characteristics of human visual system and the basic characteristic of the image,the paper puts forward two kinds of image quality assessment methods, its main content can be summarizes as follows:(1) We proposed a 2D image quality assessment method based on visual attention. The image quality assessment algorithm is based on visual information fidelity, physiology and psychology characteristics of the human visual perception, fully considering the visual attention mechanism, and different weights were given to different regions, after that the fidelity of distortion image and original image from the viewpoint of information theory are calculated. The experiment results proved that the consistency of the proposed method and subjective assessment results is excellent.(2) We proposed a 3D image quality assessment method based on threshold segmentation and feature extraction. The metric has taken into account visual attention to remove uninteresting information through setting visual threshold, and extract the image feature by the classical canonical correlation analysis algorithm. Then establish the proposed metric by combing the calculated feature similarity with the feature of depth information. The experimental results show that the algorithm can reflect the human subjective feeling of 3D image quality very well.