PERCEPTUAL QUALITY OF ASYMMETRICALLY DISTORTED STEREOSCOPIC IMAGES : THE ROLE OF IMAGE DISTORTION TYPES

Objective quality assessment of asymmetrically distorted stereoscopic 3D images is a challenging problem, for which direct simple combinations of the quality of the left and right views fail to produce adequate predictions. We carried out a subjective quality assessment experiment on a database that contains both single-view images and stereoscopic images with symmetric and asymmetric distortion types and levels, where the distortion types include noise contamination, blur and JPEG compression. Our results suggest that simply averaging the quality of the left and right view images well predicts the quality of symmetrically distorted 3D images, but generates substantial bias when applied to asymmetrically distorted stereoscopic images. More interestingly, we find that such bias could lean towards opposite directions, largely depending on the distortion types. We propose a computational model that accounts for the bias, leading to significantly improved quality prediction of stereoscopic images.