Study On Visual Perception Based Screen Content Image Quality Assessment Method

With the rapid development of the cloud computing and multimedia communication technologies,screen content images(SCIs)have been widely used in remote computing,virtual desktop sharing and online education,etc.,which has attracted more and more attentions from both academia and industry communities and become a hot research topic in the field of image processing.SCI is a mixture of natural images and computer generated graphical content,such as icons,charts and characters.Similar to the natural images,SCI is inevitably involved distortions,such as,Gaussian noise,quantization distortion,and contrast distortion,during the process of generation,transmission and displaying,leading to the quality reduction of SCI.However,due to the distinctly different statistical characteristics of SCI and natural images,the traditional natural image quality assessment method is not directly applicable to SCI.Therefore,it is of great theoretical significance and practical application to accurately evaluate the quality of the SCI based on the human visual system(HVS)and the features of SCI.This thesis mainly considers the characteristics of HVS and SCI to develop the visual perception based SCI quality assessment methods and proposes three SCI quality evaluation methods,respectively.They are the screen content image quality assessment using edge information,the screen content image quality assessment using multi-scale difference of Gaussian,the screen content image quality assessment using radial symmetry.The details of these methods can be summarized as follows:1.Considering that the HVS is more sensitive to edge information and the SCI include a large number of edge information,a SCI quality assessment method using edge information is proposed.Firstly,the proposed method individually extracts the edge information of the reference SCI and its distorted version from the spatial and frequency domains and further combines them to compute the edge information similarity map.Then,the interest region in the SCI is established based on the edge information,which is used as a weighting map to combine the similarity map and get the quality score of the distorted SCI.2.Since the SCIs contain a lot of symmetrical information,such as eyes,nose and ears in the pictorial area,letters and graphic in the textual area,the symmetry information of SCI is extracted by using radial symmetry and used to reflect the quality degradation of SCI.Firstly,we convert RGB color channels to LMN color channels.Then,the radial symmetry similarity of the input reference SCIs and the distorted SCIs is calculated based on the magnitude map and the direction map,respectively.Meanwhile,the similarity of the chrominance in the reference SCI and distorted SCI is calculated.Finally,the magnitude-based radial symmetry information is used as a weighting strategy to obtain the final quality score of the distorted SCI.3.Considering that difference of Gaussian(DoG)can imitate how the human retina extracts the image details,and the multi-scale can capture the image details more effectively,we propose a novel screen image quality assessment method based on multi-scale DoG.Firstly,the proposed algorithm calculates the different scale edge map of the input image using multi-scale DoG,that is,the small-scale edge map and the large-scale edge map.The edge similarity map is generated in terms of the smaller-scale edge map,the weighting map of the input image is calculated based on the larger-scale edge.Then,the objective quality score of the distorted SCI is obtained by using the weighting map to combine the edge similarity map.In summary,this thesis presents three SCI quality assessment methods by exploring the characteristics of the SCI and the HVS.Experimental results indicate that all the proposed algorithms can accurately describe the subjective perception of SCI.To a certain extent,this thesis have achieved a certain innovation in theory,opened up new ideas for engineering implementation,and promoted the research of SCI quality assessment in terms of visual perception.