Visual comfort and depth perception measurement for stereoscopic image retargeting quality assessment

Most stereoscopic image retargeting quality assessment (SIRQA) algorithms ignore the binocular difference between the left and right views on visually important content and the relative depth difference between the original and resized images, lowering the performance of the SIRQA algorithms. To address these issues, we propose a metric to measure the visual comfort of stereoscopic retargeted images, which assesses the binocular inconsistency caused by the difference between the left and right views in terms of the matched pixel pairs and information loss in salient regions. We also present a metric to evaluate the depth perception distortion of stereoscopic retargeted images, which calculates the relative depth between the background and the foreground objects in the original and the retargeted image respectively, and measures the relative depth difference between the original and the resized photos. Furthermore, we adopt the two proposed metrics to a SIRQA framework based on image classification to perform the quality evaluation of the stereoscopic resized images with other metrics. Experimental results demonstrate that the performance of the proposed SIRQA method outperforms the state-of-the-art algorithms. Moreover, ablation studies indicate that the proposed metrics can effectively improve the consistency between subjective and objective evaluations.