The Study On 3D Image Quality Assessment Based On Block Parallax And Saliency

In recent years,the output of stereoscopic image content and the consumption volume have ushered in explosive growth.The current mainstream 3D stereoscopic display technology brings depth perception to the viewers by projecting the images with parallax acquired by dual viewpoints onto the left and right eyes.However,due to the difference with real 3D perception,3D images also cause viewers to experience a series of discomfort and exhaustion,including: headache,dizziness,nervousness,tears,and eye astringency.Therefore,the 3D image quality assessment method that is consistent with the subjective experience is particularly important.Since the left and right viewpoint images are displayed on the fixed plane different from the actual 3D perception during the viewing,the parallaxes of the left and right viewpoints lead to different degrees of confusing-adjustment contradiction by the human eye.On the other hand,the human visual system has a selective attention mechanism.The quality evaluation model based on image saliency has also become a research hotspot in recent years.How to extract relevant features based on binocular visual differences and how to use image saliency for constructing stereoscopic image quality assessment models is the main content of this paper.In order to measure the binocular visual difference,this paper studies a block parallax feature that can be used for 3D image quality assessment,and constructs a related stereoscopic image quality evaluation model using superpixel segmentation.The block parallax characterizes the difference in the same position area of the left and right views of the screen,and its physiological basis is the receptive field mechanism and binocular vision.In the actual calculation,we give a variety of strategies based on the block method,the block size,and the difference measurement method.In this paper,the left and right graphs are superpixel-blocked based on pixel similarity,and the left and right graph differences are calculated by using the histogram absolute difference and the histogram K-L distance respectively to obtain the block disparity map.Finally,the relevant statistical characteristics of the block disparity map are used as input.Construct SVR regression prediction model.The experimental results show that the block parallax quality evaluation method described in this paper can be used to model the human visual system better,and the performance is good;the quality evaluation model of the superpixel segmentation strategy is better than the quality evaluation model of the rectangular segmentation strategy.In order to explore the influence of image saliency on the block parallax quality evaluation method,this paper designs a 3D block saliency map extraction method based on block disparity map,proposes a time-series perception model based on image segmentation,and uses 3D block saliency map for Gain quality evaluation model constructed from rectangular block parallax.3D significance includes two aspects of 2D significance and binocular visual significance.In this paper,the 2D saliency map is extracted by the GBVS(Graph-based visual saliency)method,and the block disparity saliency map is obtained by linearly fusing the block disparity map into a block depth map and a block depth map gradient map.Finally,the above two saliency maps are linearly fused to construct a 3D saliency map.On the other hand,we see the process of human eye observation as the process of fixation,and different fixation points represent different receptive fields.At the same time,the level of significance means the length of time that the human eye looks at the area.Based on the above understanding,this paper proposes a time-sense model based on image segmentation,which maps the image saliency size to the frequency of the retinal response of the component block image in the human eye.Finally,based on the understanding of the temporal awareness model,the 3D block saliency map is weighted to construct a three-dimensional image quality evaluation model.The experimental results show that the weighted model has a gain relative to the original block parallax quality evaluation model,indicating that the image saliency has an impact on human eye comfort perception.The temporal perception model proposed in this paper is more consistent with the human eye perception process.