Binocular just noticeable difference based stereoscopic image quality assessment

摘要

Based on human eyes' visual perception, a new stereoscopic image quality assessment (SIQA) metric is proposed in this paper, which brings binocular masking phenomenon to SIQA with a binocular just-noticeable difference (BJND) model. This paper also presents a new distortion evaluation model called contrast-structure-gradient similarity (CSGS) model. The distortion assessment and binocular characteristics are fused to estimate the objective quality of left and right views. Then an appropriate combination of the two views is used to get the overall quality of stereoscopic image pair. For symmetric distorted images, the quality of a stereoscopic image is the average quality of the two views; but for asymmetric distorted images, the weights of the two views are different, because the contribution of the two views to human visual perception are different for different distortion types. Experimental results show that the proposed metric outperform the relevant existing metrics.