Blind 3D image quality assessment based on self-similarity of binocular features

The perceived quality assessment of three-dimensional (3D) images has emerged as a challenging research topic in the field of 3D imaging in recent years. Especially, blind quality assessment of 3D images encounters challenges due to prior information about the original 3D images is not available. In this paper, we propose a blind 3D image quality assessment (IQA) metric that utilizes the self-similarity of binocular features. The primary contribution of this study is that the proposed metric considers the binocular visual property and the local visual structural property for blind 3D-IQA. We calculate the self-similarity of binocular rivalry response as well as binocular orientation selectivity in the distorted 3D image. We then extract the inter- and intra-pixel binocular quality-predictive features from these self-similarity measures. Following feature extraction, we use machine learning based on the support vector regression (SVR) procedure to drive the overall quality score. Our results on publicly accessible 3D databases confirmed that the proposed metric is highly efficient and robust.