Blind Image Quality Assessment Algorithm Based On The Perception Characteristics Of Human Visual System

Blind image quality assessment(BIQA)is a significant research task in the field of image processing,which aims to design a computing model that does not rely on any prior knowledge and can automatically evaluate image quality.Its works quantify the performance of image and provide an important basis for other image processing algorithms.This paper takes natural image as the research object,and carries out the following research work for BIQA.(1)Existing blind image quality assessment neglecting the characteristics of human visual perception,a blind image quality assessment algorithm based on dual-channel deep network with enhanced visual sensitivity is proposed.The algorithm combines visual sensitivity with deep learning,extracts high-level visual perception features and deep features from JND map and distorted image respectively,and further integrates them.To further improve the performance of the algorithm,a new loss function is proposed to enable the network to learn the deep image quality perception features by evaluating the quality rank between different distorted images during the learning process.The experimental results show that the correlation between the evaluation results of the algorithm and the subjective evaluation is 0.98 on the LIVE dataset,0.88 on the LIVE Challenge and 0.88 on the TID2013,respectively.(2)The performance of BIQA method based on deep network is usually limited by dataset,which leads to the low universality of the model.Therefore,this paper proposes a BIQA(unsupervised)algorithm based on visual saliency and multi feature fusion.By studying the characteristics of primary visual perception,this algorithm designs a new feature descriptor to express image contrast and hue distortion.At the same time,it combines the multi-level natural scene statistical features,structural features and color statistical features of the image to express image distortion more comprehensively.Then,based on the mechanism of visual significance,it establishes an image quality calculation model to predict image quality.The experimental results show that the correlation between the evaluation results of the algorithm and the subjective evaluation is 0.93 on the LIVE dataset,0.54 on the LIVE Challenge and 0.48 on the TID2013,respectively.Compared with the other DU-OU BIQA algorithms,the accuracy and universality were further improved.(3)Finally,the proposed IQA algorithm in this paper is applied to image denoising algorithm BM3 D.A noise estimation algorithm is constructed by combining the image quality-aware features to guide BM3 D to automatically obtain the noise variance parameters.Experimental results show that the proposed algorithm can effectively improve the performance of BM3 D denoising algorithm.In conclusion,the blind image quality evaluation algorithm proposed in this paper can achieve a highly effect consistent with the subjective evaluation of human vision.At the same time,the algorithm can be used to guide the research in other areas of image processing.