3D Image Quality Assessment Based On Binocular Visual Characteristic

Stereoscopic images are affected by a series of factors such as acquisition,transmission,processing,storage and compression,resulting in image's distortion on the terminal display.The distortion of image seriously hinders that people understanding the world.In the field of image processing,how to effectively evaluate stereo image quality has become a challenging research topic.In recent years,there has been new progress in the research of human visual systems in the field of biology,which has promoted the study of models of stereoscopic image quality evaluation on simulating binocular visual characteristics.We summarized research findings of predecessor,and we propose two new objective methods for stereo image quality evaluation based on binocular visual characteristics.Firstly,we propose a stereoscopic image quality evaluation method based on wavelet transform and structural features.The different channels of the human visual system are similar to the multi-channel filtering of wavelets.Compared with the traditional wavelet transform,the dual-tree complex wavelet transform has some advantages such as nearly shift invariance and more directional selectivity.We decompose the wavelet into three levels,and six directions per level.First,we perform double-tree complex wavelet transform on the stereo image,disparity map and cyclopean map,and then uses the gradient amplitude and the phase amplitude characteristic of the Log Gabor filter to describe the structure of the wavelet sub-band.The experimental results on two distorted stereo image databases(LIVE-3D I and LIVE-3D II)show that the result of the proposed algorithm is highly correlated with the subjective evaluation result.Secondly,we propose a stereo image quality evaluation method based on grid characteristic and Weibull statistical characteristic.In recent years,the shape parameters and scale parameters of Weibull distribution have a strong correlation with human visual perception on image,which has become a research hotspot in the field of image quality evaluation.Grid characteristic can describe image structures.However,the grid features are not sensitive to image's WN distortion,and the shape and scale parameters of the image does not effectively describe the effect of structural distortion of image.In order to overcome these difficulties,we combine the extracted grid features of the stereo image and the Weibull statistical features to compensate for each other and improve the evaluation effect.The experimental results on the databases LIVE-3D I and LIVE-3D II show that the evaluation results obtained by the proposed method are highly consistent with the subjective scores.