Visual comfort assessment for stereoscopic images based on sparse coding with multi-scale dictionaries

Predicting the experienced visual comfort in line with human subjective perception is a non-trivial task in stereoscopic three-dimensional (3D) imaging system. For the task of 3D visual comfort assessment (VCA), an intuitive idea is to develop effective subjective assessment-like models. As one of the most widely used subjective assessment methods, the absolute category rating (ACR) requires each observer to conduct a multi-grade judgment with respect to one's subjective sensation of visual comfort. Motivated by this fact, we propose a novel VCA method for stereoscopic images based on sparse coding with multi-scale dictionaries. The main technical innovations of our proposed method are threefold. First, we construct multi-scale dictionaries that correspond to the multiple visual comfort scales in subjective assessment to serve as the prior knowledge for objective VCA. Second, we apply the sparse coding algorithm to estimate a testing sample's probabilities of belonging to each scale and also to derive multiple scale-specific visual comfort scores. Finally, by linearly combining all the scale-specific visual comfort scores with their corresponding probabilities as weights, the final visual comfort score is generated. Experimental results on two benchmark databases demonstrate that the proposed method can achieve higher consistency with human subjective perception compared with some state-of-the-art approaches.