Research On Panoramic Image/Video Quality Assessment

The rise of virtual reality has provided users with different visual experiences from the past.As an important part of virtual reality technology,panoramic image / video has become a research hotspot in the field of image processing.However,with the continuous development and popularization of virtual reality technology,the demand of user for high-quality immersive visual experience is also increasing.However,in the process of splicing,projection,compression,transmission,storage,etc.,it will inevitably bring damage to the quality of panoramic images and videos.Therefore,the research work on the evaluation of panoramic images and videos is crucial.We focus on the study of the panoramic image/video quality assessment algorithms;the main research contents are as follows:Firstly,a full-reference panoramic image quality assessment algorithm based on phase consistency is proposed.The algorithm firstly weights the planar image panoramically,so that the features on the plane can accurately reflect the spherical spatial quality distortion.Then we uses the mutual information of phase consistency to obtain the structural similarity between the reference image and the distorted image,and the similarity of the phase consistency local entropy is used to reflect the similarity between the texture of the reference image and the distorted image,and the two parts of similarity are fused to obtain the objective quality score of the panoramic image.Experimental results show that the proposed algorithm can obtain a higher consistency with the subjective score.Next,a full-reference panoramic video quality assessment algorithm based on statistical similarity is proposed.The algorithm first finds that the statistical distribution of the single plane of the source video frame in the Cubic Mapping Projection is close to the generalized Gaussian model,and the distortion will destroy this statistical law,and then the distorted video frame is corrected by the panoramic weight to make it conform to the generalized Gaussian distribution.On this basis,the similarity of the description parameters of the statistical distribution between the reference video and the distorted video is used as the panoramic feature to evaluate the quality of the distorted video.Experimental results show that the proposed algorithm is superior to other mainstream algorithms in panoramic video and image data setsThen,a no-reference panoramic image quality assessment algorithm based on the asymmetric mechanism of the human brain is proposed.According to the asymmetric mechanism of the left and right hemisphere when the human brain processes visual information,the panoramic high-frequency feature and the panoramic low-frequency feature are extracted from the panoramic image respectively.The gray level cooccurrence matrix extracts panoramic high-frequency features,and then uses panoramic weighted luminance histograms to represent panoramic low-frequency features.Finally,SVR is used to build a quality prediction model from feature space to quality score space.Experimental results prove the superiority of the fusion feature.Finally,a no-reference super-pixel-based panoramic video quality assessment algorithm is proposed.The algorithm first converts pixel-level images into superpixel images,and then proposes a superpixel-based panoramic weight by combining projection format and human perception.The fused panoramic weights can map the planar structural features into spherical panoramic structure features more accurately.Finally,the panoramic structure feature is used as the input of the SVR to obtain the panoramic video quality score.Experiments prove that the proposed algorithm achieves excellent performance on both panoramic video and image datasets and is particularly robust against different types of distortion.