Research On Image/Video Objective Quality Assessment Algorithms Based On Contrastive Principal Component Analysis

With the rapid development of communication networks,a large number of high-resolution and high-quality images and videos are transmitted in communication networks.How to accurately evaluate the perceived quality of images or videos is one of the hot topics in the field of image processing.The current image quality assessment algorithms still have many problems,such as low accuracy,weak generalization ability and few applicable scenes.To solve these problems,this paper proposes a series of image/video objective quality assessment algorithms based on contrastive principal component analysis.The research work and results of this paper are summarized as follows:(1)A simple,fast and effective full-reference image quality assessment algorithm ICPCA based on contrastive principal component analysis is proposed.Firstly,the multi-scale contrastive principal component analysis algorithm is used to extract the principal components from the reference image,and the contrastive principal components are extracted from the reference image and the distorted image respectively,and the correlation coefficient between them is calculated as the feature.Then by training BP neural network,the correlation coefficient is fitted to the mean opinion score of the picture.The PLCC of the proposed ICPCA algorithm on CSIQ and LIVE 3D IQA reaches 0.9309 and 0.9320,respectively.(2)ICPCA is extended to video quality assessment by using different temporal pooling strategies and named VCPCA.The video is extracted as an image frame by frame,and the well-trained network generated by TID2013 dataset is used to calculate the frame level quality of the video,and then the final score of the video is obtained by using different temporal pooling strategies according to the different videos.Improvement of VCPCA obtained through temporal feature extraction,T-VCPCA.The PLCC of T-VCPCA on MCL-V,a video quality assessment dataset,reaches 0.9351,and the PLCC under H.264 compression distortion reaches 0.9556.(3)ICPCA is applied to virtual reality image quality assessment.Experiments were carried out on LIVE 3D VR IQA and CVIQ,and the PLCC of ICPCA on CVIQ reached 0.9499.Experiments across datasets show that LIVE 3D VR IQA has better generalization ability than CVIQ-trained networks.To sum up,this paper proposes a series of image/video quality assessment algorithms based on contrastive principal component analysis.The proposed algorithms have a particularly prominent effect on the compression distortion of images and videos,and also shows strong adaptability in the application of Virtual Reality scenes,which solves many problems such as low accuracy,weak generalization ability and few applicable scenes of traditional image quality evaluation algorithms.