Full-reference quality assessment of stereoscopic images using disparity-gradient-phase similarity
Li Fei-Fei, Liquan Shen, Dapeng Oliver Wu, Ruigang Fang · 2015
With the effort done by many researchers on 2D image quality assessment, the correlation between subjectivity scores in the databases and the objective scores obtained by current image quality metrics reaches to a high level. However, stereoscopic image quality metrics do not achieve a good performance on 3D image databases. It's important to take advantage of the high efficient 2D image quality metrics as well as the properties of stereoscopic images to form a new metric. In this paper, we develop a novel full-reference quality metric for stereoscopic images using disparity-gradient-phase-congruency similarity (DGP-SIM). Firstly, cyclopean images are generated using the stereo pairs by considering the binocular perception and combination properties of human visual system (HVS). Secondly, phase congruency (PC) and gradient magnitude (GM) are extracted from the original and distorted cyclopean images, and then two similarity indexes are computed. We also compute the disparity similarity which can influence stereoscopic image quality. Finally, three feature similarity indexes are mixed together to derive a single quality score. Experimental results show that the proposed metric can achieve a good consistency with subjective assessment of stereoscopic images.