Research On Stereoscopic Image Perceptual Quality Assessment For Three-dimensional Video System

The quality of image and visual security are the top concerns while the viewing of a stereoscopic image is conducted. The main work of this paper is the research of perception quality assessment of stereoscopic image from two perspectives: stereoscopic image quality assessment(SIQA) and visual comfort assessment(VCA). The major contents are as follows:1) Binocular just noticeable difference(BJND) based SIQA metric. BJND model is used to describe the property of binocular masking effect quantitatively and a structural similarity model is proposed to evaluate the degree of distortion. By fusing outputs of two models, perceptual quality of each view is obtained. Then an appropriate combination of two views is used to acquire the overall quality. Experimental results show that proposed metric outperforms the relevant existing metrics. For symmetric and asymmetric distortion, the Pearson Linear Correlation Coefficient(PLCC) is over 0.94 and 0.91 respectively.2) For a given image with preliminarily-known distortion type, perceptual quality will be predicted more precisely. Therefore, on the basis of analyzing characteristics from distorted images, we propose two metrics.(1) 2D(2 Dimensional) metric. A threshold-based classifier is built to ascertain the distortion type. And three different models are utilized to evaluate corresponding three types of distorted images. Experimental results show the classification accuracy reaches 100% on the condition of only three distortion types are involved.(2) 3D(3 Dimensional) metric. A new SVM(Support Vector Machine) based multi-classifier makes the distorted images classified. And the quality of different images is predicted adaptively by a proposed binocular fusion based assessment metric. Experimental results show the accuracy of classification exceeds 99% and 93% on intra-database and inter-database respectively. Besides, objective quality predicted by these two metrices has a high consistency with subjective perception.3) Two VCA methods are proposed from the perspectives of objectivity and subjectivity respectively.(1) Objectivity: For each stereoscopic image, extract disparity, width and complexity from its foreground object as features. Considering the psychological phenomenon that object’s width and complexity do not consistently influence the perception of visual comfort, the whole images are divided into four categories and four corresponding assessment models are exerted. Experimental results show that proposed metric outperform existing metrics, and PLCC exceeds 0.84.(2) Subjectivity: A “double stimulus” based subjective experiment is made to find the disparity which can cause just noticeable discomfort, and formulate the relationship between luminance distance and subjective visual comfort score. Results show that disparity within 1° and exceeds 2° indicates comfort and extremely discomfort respectively. What’s more, the relationship between visual comfort and luminance distance is arrived as monotonicity and linearity.