Stereoscopic Video Quality Assessment Based On Binocular Perception

In the information age,with the rapid development of 3D technology,the explosive growth of the demand for stereoscopic videos comes.Meanwhile,it is a great challenge for stereoscopic video providers to ensure high quality viewing experience.Therefore,this paper focuses on the research of stereoscopic video quality evaluation,and the research is carried out from subjective and objective two aspects.In the subjective evaluation,this paper establishes a relatively complete 3D video quality assessment database,which contains a large number of video samples,rich video content,mixed-resolutions,and introduces both symmetrically and asymmetrically distortions.This database is more in line with actual needs,and is freely available,which is conducive to the improvement of stereo video evaluation algorithms.In the objective evaluation,we propose a blind multi-view feature learning metric,which jointly focuses on the spatial,temporal and spatio-temporal interaction information.Firstly,binocular summation image and difference image are extracted as spatial views,and a set of local binary patterns(LBP)statistical features are used as spatial description.Secondly,optical flow statistical features are used to describe the temporal distortion.Thirdly,the inter-frame difference is computed as spatio-temporal interaction view,and the same as spatial view,LBP features are used to describe spatio-temporal distortion.Subsequently,a support vector regression(SVR)is utilized to map the feature vectors of each single view to subjective scores.Finally,the scores of multiple views are pooled into the final score according to their contribution rate.In this paper,we first establish a novel stereo image quality assessment model and a novel stereo matching model.Then we establish a relatively complete and freely available stereoscopic video quality assessment database.And by jointly focuses on the spatial,temporal and spatio-temporal interaction information,we put forward an accurate and effective objective evaluation method.Experimental results on three stereoscopic video databases verify the effectiveness of the proposed method.