User Demands Oriented Visual Information Supply Method

Digital images are one of the main carriers for people to express information at present.While a large amount of digital image information enriches our lives,it also brings great challenges to network bandwidth and storage capacity.To this end,in the process of image transmission,storage,etc.,the redundant information in the image is often reduced by image encode.However,this approach usually reduces the quality of the image,and the image quality after attenuation may not meet the specific needs of users.Therefore,a flexible strategy is needed to match the different needs of the user with the corresponding encoded image quality.This thesis analyzes the potential application scenarios that users may have,and divides the quality of the image in a demand-oriented manner.Based on the mechanism of the human visual system,a full-reference image quality assessment algorithm and a just noticeable distortion algorithm are designed to measure the quality of images.Furthermore,a demand-oriented image supply program is realized.The research contents of this thesis are as follows:Firstly,three types of requirements are divided for different cognitive purposes: considering the existence of special occasions that require the original and complete information of the image,an absolutely lossless requirement that uses lossless compression processing is set;for the cases that do not require the original information of the image but require the degradation of image quality is invisible to the human eye,the perceptively lossless requirement is set;the visual system has limited ability to discriminate between different noise levels,and for the needs of different levels of image quality,there are four levels of cognitively lossless requirements.Then,the cognitively lossless requirements are measured based on image quality level.Gradient,complexity,and anisotropy are used to systematically decompose the image into five regions with different properties,including primary edges,secondary edges,complex textures,regular textures,and smooth regions.We deeply analyze the different effects of noise in these areas,and the distortion of gradients,phases,and other features in different regions is measured differently,so a full reference image quality assessment algorithm based on image structure decomposition is proposed to achieve different levels of cognitively lossless requirements based on lossy compression coding.Finally,with the help of image quality assessment constraints,the first just noticeable distortion threshold calculation models based on unsupervised learning is proposed.Due to the lack of pixel-level threshold label,it is impossible to directly construct a supervised deep learning networks to achieve the just noticeable distortion threshold prediction.To this end,the correlation between image quality assessment and just noticeable distortion is analyzed in depth,and the just noticeable distortion is mapped back to the image with the help of noise patterns,and the network parameters are optimized with the constraint of high quality to achieve unsupervised image just noticeable distortion threshold estimation.Besides,it achieves the measurement of perceptively lossless requirement based on lossy compression coding.Based on the above theories,with the help of the proposed image quality assessment and the just noticeable distortion,an image content supply system oriented to user demands is designed.The system can accurately provide images of corresponding quality according to the user's cognitive needs,and minimize redundant information under the condition of satisfying the cognition.