Research On Image Quality Assessment Method For Interactive Three-Dimensional Video System

Interactive three-dimensional video system with features such as stereoscopic perceptual and interactive has gradually become the key research domain of next-generation multimedia systems. Because image quality is an important indicator of the performance of interactive three-dimensional video system, image quality assessment has been one of the key technologies in this field. This dissertation combines the characteristics of the human visual system and single view image quality evaluation method to study the image quality analysis method in interactive three-dimensional video system. The main contributions include:(1) Focusing on the features of transmission link in the interactive three-dimensional video system, we discuss the features and applicability of objective model of single view image quality evaluation. Then, a novel reduced reference metric based on Contourlet transform is proposed. After investigating of the statistical properties of Contourlet transform domain coefficients, a new reduced reference metric is proposed. Experimental results show that compared with the previous reduced reference assessment algorithm, these two proposed algorithms are more accurately when evaluate the image quality. And the correlation between these two metrics and subjective evaluation scores is higher than the traditional evaluation metric such as PSNR.(2) Three subjective perception experiments are designed after discussing the characteristics and scope of application of a variety of subjective and experimental standard test methods. Experiment I establishes a stereoscopic image database, containing 10 pairs of the original stereoscopic images of four kinds of distortion types. Total 370 distorted stereoscopic images have been evaluated by participants in subjective experiments. Experiments results show that the relationships between distortion parameters and perceived quality of stereoscopic images are different for distortion types. Human rating on quality of stereoscopic image will be insensitive, when the quality of left view image keep constant and the right view is Gaussian blurred. Quality rating of stereoscopic image will be the most sensitive, when the quality of left view image keep constant and the right view is JPEG compressed. Experiment II investigates how low-pass filtering on JPEG compressed stereoscopic image can enhance the stereoscopic perceptual quality. The results of experiments II show that Gaussian blurring may be used to enhance the perceived quality of stereoscopic images in which the right view image is JPEG compressed at an extent of distortion. Furthermore, the enhancement is image content dependent. Experiment III investigates quantitative research methods of masking effect in binocular visual perception. The just noticeable differences (JND) of asymmetric stereoscopic video compression are studied and the JNDs on natural image content by human eyes under general viewing conditions are evaluated. Paired comparison method is used for subjective visual experiment. The results indicate that JND value is image content and quality dependent.(3) Two stereoscopic image quality objective metrics are proposed based on the subjective perception experiment research. Experiment results show that the two proposed metrics with low computation complex has good consistency with human subjective perception.