Universal blind image quality assessment for stereoscopic images

Quality assessment of stereoscopic 3D images is a challenging field and represents a key factor in the success of 3D multimedia applications. Despite the important research effort in the last few years, there is no commonly accepted metric ensuring a reliable 3D quality evaluation. This statement becomes even worse when it comes to asymmetrically distorted stereoscopic content. In this paper, we propose a universal blind quality assessment metric for stereoscopic images relying on: 1) distortion type detection and 2) asymmetry nature determination. Based of the latter key information, the 3D image quality is appropriately estimated using a binocular combination strategy. Experimental results showed that the proposed metric reached a significant prediction consistency and accuracy when compared with state-of-the-art metrics.