Reduced Reference Stereoscopic Image Quality Assessment Based on Binocular Perceptual Information

Abstract

In this paper, we propose a novel reduced reference stereoscopic image quality assessment (RR-SIQA) metric by using binocular perceptual information (BPI). BPI is represented by the distribution statistics of visual primitives in left and right views’ images, which are extracted by sparse coding and representation . Specifically, entropy of the left view’s image and entropy of the right view’s image are used to represent monocular cue. Their mutual information is used to represent binocular cue. Constructively, we represent BPI as three numerical indicators . The difference of the original and distorted images’ BPIs is taken as perceptual loss vector. The perceptual loss vector is used to compute the quality score for a stereoscopic image by a prediction function which is trained using support vector regression (SVR). Experimental results show that the proposed metric achieves significantly higher prediction accuracy than the state-of-the-art reduced reference SIQA methods and better than several state-of-the-art full reference SIQA methods on the LIVE phase II asymmetric databases.