Visual Perception Based Blind Stereoscopic Omnidirectional Video Quality Assessment

Stereoscopic omnidirectional videos (SOVs) can provide users with continuous immersive visual experience, but the distortions introduced in the process of its processing, coding transmission, visualization bring great challenges to its quality assessment. In this paper, a blind SOV quality assessment method including spatial perception model (SPM) and temporal perception model (TPM) is designed based on the spatio-temporal separability of neurons in the region 18 of visual cortex, and the characteristics of stereoscopic and omnidirectional perception are considered. In the SPM, the fractal dimension is introduced to design the pixel domain feature extraction module of single viewpoint; based on the theory of binocular summation and binocular difference to regulate binocular behavior, the two-stage gain control and maximum response model are combined to construct the binocular perception model, and the orientation feature in transform domain of the binocular perception map is extracted to achieve the complementary role of content perception of single viewpoint and binocular. For the TPM, considering the motion perception of the middle temporal region and the correlation between the left and right viewpoints, a binary statistical model for temporal information extraction is constructed to assist SPM to form functional complementarity. Experimental results show that the proposed method has good quality assessment performance, and has a better consistency with human visual perception.