Blind stereoscopic 3D image quality assessment via analysis of naturalness, structure, and binocular asymmetry

Over recent years, stereoscopic three dimensional (S3D) images have grown explosively and received increasing attention. Quality assessment, as the fundamental problem, plays an important role in promoting the prevalence of S3D images as well as the associated products. In this paper, an effective blind quality assessment method of S3D images is proposed via analysis of naturalness, structure, and binocular asymmetry. To be specific, given that natural images obey certain regular statistical properties, natural scene statistic (NSS) features of left and right views are first extracted to quantify the naturalness. Second, by considering binocular visual characteristics, statistical features are extracted from a created cyclopean map. Moreover, gray level co-occurrence matrix (GLCM) is utilized to capture quality-sensitive features from the cyclopean phase map. Third, to quantify the asymmetric distortion, a simple but effective measurement is utilized, i.e., calculating the similarity between left and right views as well as statistical features of their difference map. Finally, all extracted quality-sensitive features are combined, and trained together with the subjective ratings to form a regression model using support vector regression (SVR). Experimental results on four publicly available databases (two symmetrically distorted databases and two asymmetrically distorted databases) demonstrate that the proposed method is superior to several mainstream image quality assessment (IQA) metrics.