Stereoscopic Image Quality Assessment Based On Binocular Rivalry

Nowadays,the emergence of 3D technology gradually changes our lifestyle,however,measuring the quality of 3D multimedia(images and videos)is still a challenge.In this paper,we propose three Stereoscopic Image Quality Assessment(SIQA)frameworks that focus on both the innovation of binocular visual properties and distortion-aware image features.To achieve this goal,we firstly propose a Full-Reference(FR)SIQA model accounting for binocular rivalry.By fully considering binocular rivalry phenomena,we try to simulate the physiological process of cyclopean image.Meanwhile,image low-level features(local phase and local amplitude)are employed to describe visual degradation.The final quality score is computed by comparing the similarity of phase and amplitude between reference and distorted stereopairs.Extensive experiments on LIVE 3D Image Databases demonstrate that the proposed framework achieves higher consistency with subjective tests than relevant SIQA metrics.And,we further discussed the different roles of amplitude and phase in determining image quality.In addition,to overcome the disadvantages of FR SIQA methods,we proposed a blind SIQA model according to binocular summation and difference channels.In this method,natural scene statistics extracted from the two channels are used to describe visual degradation.At last,another reduced-reference method is also proposed to explore the applications of sparse representation in SIQA.In this model,we combine the stereo vision with sparse coding,and extract useful visual features from four different channels.Experiments further verify the rationality of the proposed method.