Research On Visual Comfort Assessment Based On Visual Features Of Stereoscopic Content

It is known that the prefect3D visual comfort experience wouldprovide viewers the truly immersive experience and various depth details.But based on two-viewpoint3D rendering technique, it cannot providesuch perfect visual experience. It will not only induce the viewer’s visualfatigue, but also provide false various stereoscopic visual details. So in ourthesis, we try to analyze the visual features of3D scenes, and to establishthe effective visual comfort model based on3D scenes visual feature vectorcombination and subjective visual comfort test data.In our thesis, we mainly focus on researching and analyzing visualcomfort assessment method based on visual features combinations of3Dcontents. Specially, our main works and innovations are as follows:Firstly, we design and build up the stereoscopic image database VBEDand stereoscopic video database RVFD. In fact, building up stereoscopicimage/video database is important and meaningful. It could not onlyproviding more specific data statistics, but also give the subjective visualcomfort test data. Specifically, VBED database is to represent the variousvisual features of border objects in3D content. It is mainly for the study ofthe correlation between border effect and visual comfort. In previousacademic researches, no such database is to analyze the relationshipbetween the border effect and visual comfort. Thus our database VBED istrying to fill such gap. Similarly, RVFD database is to represent the variousvisual features of Regions of Interest in3D content. It is mainly for the studyof correlation between3D video content visual features and visual comfort.Secondly, we try to analyze and research the correlation between visual features of stereoscopic image content and visual comfort. Forstereoscopic image content, previous researchers mainly focus on objectsin center regions of content. Few may focus on border effect and visualfeatures of border objects. In our thesis, we try to analyze and assess thecorrelation between border effect and visual comfort. Thus we propose thevisual comfort assessment based on visual features of border objects in3Dimage content. In detail, first of all, we analyze visual features combinationsof border objects, such as disparity features vector, thickness featuresvector, blur features vector and color contrast features vector. Secondly, weutilize stereoscopic image database VEBD to obtain subjective visualcomfort test data. Finally, we build the visual comfort objective evaluationmodel based on support vector regression algorithm. By analyzing thecorrelation between subjective and objective results, it shows that ourproposed visual comfort model could achieve good performanceevaluation.Finally, we try to analyze and research the relationship between visualfeatures combinations of3D video content and visual comfort. To betterassess above relationship, we propose the visual comfort assessment basedon visual features of regions of interest in3D video content. In detail, firstof all, we propose the region of interest extraction method based on salientmap and motion map. Secondly, we try to extract and compute the variousvisual features of regions of interest, such as disparity features vectorcombination and motion features vector combination, based on extractedregions of interest in3D video content. Thirdly, we utilize stereoscopicvideo database RVFD to obtain subjective visual comfort test data. Finally,we build the visual comfort objective evaluation model based on supportvector regression algorithm. By analyzing the correlation betweensubjective and objective results, it shows that our proposed visual comfortmodel could achieve good performance evaluation.