Stereo Image Quality Assessment Based On Stereo Visual Saliency

The quality of stereo image will decline inevitably during the process of acquisition,compression,storage,transmission,and display,further affect the stereo feeling of the human.Therefore,it is one of the main problems to be solved in the field of stereo imaging technology to determine an evaluation model of stereo image distortion.As an advanced feature of human visual system,visual salience refers to the attention distribution of the human eye to different regions of the image is different.In this paper,two methods are proposed to evaluate the quality of stereo images based on visual saliency.Firstly,the saliency maps of the right image and left image of the stereo image are obtained by GBVS,and the optimized saliency maps are obtained by considering the effect of the center bias respectively,multiply the saliency maps of the right and left image and the quality map together of the corresponding image and get quality objective assessment values the right and left image,the final objective value is obtained by averaging the two values.Furthermore,a structural similarity index measurement based on center bias(C B-SSIM)is proposed.In the second method,Stereo image quality assessment based on stereo visual saliency is proposed combined with saliency and depth information.extract 2D visual salient features,and combine it with absolute difference map to bulid the preliminary stereo vision saliency map,then optimize it by using center bias and foveal.Second ly,merge 2D image quality map and absolute difference quality map to build a stereoscopic image quality map.Finally,multiply stereo vision saliency map and stereoscopic image quality map together and get stereo image quality objective assessment value.The experimental results show that,for different types of distortion,the objective evaluation values of the proposed algorithm are highly consistent with the results of subjective evaluation.To test the performance of the proposed algorithms in this paper,stereo images of the common JPEG compression distortion,Gaussian blur distortion and Gaussian white noise distortion are selected as experiment material.The experimental results show that the quality assessment of stereo image based on GBVS and the center bias is better than that of the direct use of the central bias factor.Moreover,compared with the first method,the second method has the higher relativity with the MOS.And compared with the other objective evaluation methods in this field,the pe rformance of the proposed algorithms are significantly improved,and have higher consistency with subjective evaluation.