Research On Stereo Image Quality Evaluation Algorithm Based On Human Visual Characteristics

In Stereoscopic Image Quality Assessment(SIQA),how to effectively simulate the Human Visual System(HVS)to evaluate the image quality is an important and challenging problem.Considering the influence of human visual properties on image quality evaluation,this paper introduces the basic structure of HVS in detail,and analyzes the visual perception characteristics generated by HVS.In order to combine the human visual properties to design algorithm that is more consistent with the visual perception process of human eyes,this paper has done the following work:Firstly,this paper proposes a full reference SIQA model based on monocular and binocular visual information.Monocular and binocular visual information features are extracted from monocular images and cyclopean images respectively.Then the features are combined to obtain the final quality score.The experiment results show that the performance of the proposed method has a obvious improvement than other works.And the prediction accuracy on the LIVE 3D database is significantly improved.It also has a good performance on different distortions,which means the proposed metric has a better consistency with HVS.Then,this paper proposes a no reference SIQA model based on multi level feature fusion.The model predicts the quality of stereoscopic image through deep learning networks.This framework is composed of three single-row subnets and three layer feature fusion subnets.Three single-row subnets are used to extract different monocular and binocular features.And three layer feature fusion subnet can extract low,middle and high level visual feature information respectively.At the same time,the model maps the feature information into image quality scores through the full connection layer.The experiment results show that the proposed model has high accuracy in the evaluation of stereoscopic image quality,and has better performance than other no reference evaluation methods.