Research On Human Visual System-Oriented Screen Content Image Quality Assessment Method

With the rapid development of mobile Internet,cloud computing and Internet of Things,the screen content images(SCIs)attract more and more attention.Different from the natural images,the SCIs consist of the discontinuous-tone content generated or rendered by computer(e.g.,text,figure,chart,etc.)and the continuous-tone content captured by the camera(e.g.,natural scene image and video clip,etc.).The SCIs are also suffered from various distortions during the acquisition,storage,coding,transmission and display stages,leading a lower SCI quality.Therefore,how to accurately assess the SCI quality is a very important problem in the field of SCI technologies.Most of the existing image quality assessment(IQA)methods and databases are designed for the natural images,while the SCIs have different characteristics from the natural images;hence,they are not very suitable for the SCIs.Therefore,this thesis is to study the human visual system(HVS)-Oriented Screen content image quality assessment methods,mainly on gradient direction based SCI IQA method,edge model-based SCI IQA method,and SCI IQA database,as follows.1.By analyzing the difference of gradient magnitude of SCIs and natural images and the relationship between gradient direction and HVS,we propose a simple yet effective IQA method for SCIs based on the gradient direction.Firstly,a novel gradient direction computation is presented to extract the gradient direction that can effectively capture the distortions introduced on the SCIs.Then,by jointly considering the gradient direction similarity and gradient magnitude similarity together with the deviation pooling strategy to generate the perceptual quality score of SCIs.Experimental results have shown the proposed SCI quality metric can effectively depict the perceptual quality of the SCIs.2.By considering that the HVS is very sensitive to the edges while the SCIs contains a lot of edges,an accurate edge model-based SCI IQA is developed for assessing the SCIs.This approach uses the edge model to extract two salient edge attributes---i.e.,edge contrast and edge width,from both reference and distorted SCIs,respectively.Then,the degree of similarity of the above-mentioned two edge attributes is the measured independently,and then fused together using our proposed edge-width pooling strategy to generate the final quality index.Extensive simulation results have shown that the proposed IQA model produces higher consistency in accordance with the HVS.3.Based on the observation that the existing largest SCI IQA database has limited number and types of reference images and distortions,we construct a new SCIs database(SCID)according to the international standard ITU-R BT.500-11.This SCID contains 40 reference SCIs and 1,800 distorted SCIs under 9 different types of distortions and 5 degradation levels for each distortion type.The double-stimulus impairment scale(DSIS)method is then employed to rate the perceptual quality,and the rated scores are further processed to obtain the mean opinion score(MOS)value as the ground truth of each distorted SCI.Based on the constructed SCID,we further evaluate the performances of 13 state-of-the-art IQA methods and provide the extensive analysis.This SCID will be made available in future as an important databased and have great significance to promote the development of the SCI technologies.In summary,this thesis is to analyze and investigate the SCI image quality assessment methods based on the characteristics of SCI and human perception,and contains novelty and challenge.This thesis,in a certain extent,have opened up a new way for visual perception based SCI technologies and is of great theoretical and application value.