A Blind Stereoscopic Image Quality Evaluator With Deep Learning And 3D Visual Perception Route

Three-dimensional(3D)imaging technologies,which include processing stages such as 3D scene capture,3D compression,3D transmission,rendering,and display,have drawn tremendous research attention in the past decade.The objective quality assessment of stereoscopic images plays an important role in three-dimensional technologies.However,most of the current blind stereoscopic image quality assessment(SIQA)algorithms cannot show reliable accuracy.One reason is that they do not have the deep architectures and the other reason is that they are designed on the relatively weak biological basis,compared with findings on human visual system(HVS).This paper proposes a Deep Edge and COlor Signal INtegrity Evaluator(DE COSINE)based on the whole visual perception route from eyes to the frontal lobe,and especially focus on edge and color signal processing in retinal gangli on cells(RGC)and lateral geniculate nucleus(LGN).Specifically,the propose d DECOSINE computes two locally estimated scores: edge quality and color q uality scores.Inspired by that binocular integration occurred in V1 after edge e xtraction process of RGC,sum,difference and cyclopean maps are computed fr om Laplacian of Gaussian(LoG)filtered left and right images.The opponent c oding theory is utilized for modeling color information processing occurred in LGNs,and produced red-green(RG),blue-yellow(BY)and light-dark(Lum)m aps.From these 6 maps,quality-aware features are extracted.Furthermore,to model the complex and deep structure of the visual cortex,Segmented Stacked Auto-encoder(S-SAE)is used,which has not utilized for S IQA before.The quality-aware features are mapped into local quality scores via combination of S-SAEs/SAE and support vector regressions(SVRs).These sco res are combined into an overall score through two dynamic and one static we ighting system.The utilization of the S-SAE complements weakness of deep le arning-based SIQA metrics that require a very long training time.Experiments are conducted on the popular SIQA databases such as LIVE dat abase phase I,phase II and IVC database,and the superiority of DECOSINE i n terms of prediction accuracy is proved.The experimental results show that o ur model about the whole visual perception route and utilization of S-SAE are effective for SIQA.