Full-reference Image Quality Metric For Computer-rendered Stereoscopic Images

Abstract

In order to evaluate the quality of computer-generated stereoscopic images further accurately, we propose a new quality assessment metric based on the human visual system (HVS) modeling and machine-learning algorithm. First, we consider the existence of binocular rivalry to generate a cyclopean image of the asymmetrically distorted part that can be masked. Furthermore, we can get a pair of cyclopean images, decompose the reference and distorted images with RGB-YUV color channels respectively. We respectively calculate the edge and texture features of the 3-channel color map to obtain 6 pairs of color feature maps, then calculate the color feature difference values of each pair of feature maps. Finally, we use the support vector regression (SVR) for 6 color feature difference values to be calculated, the final quality value can be obtained. It can be seen from the experimental results that our algorithm has better performance and more accurate calculations than existing algorithms.