Research On Stereoscopic Image Quality Assessment Algorithms Based On Human Visual System

In recent years, with the rapid development of multimedia technonlogy, stereoscopic images have increasingly entensive applications. It leads to the growing demand of the clarity towards stereoscopic images. However, because of the imperfect processing technology of stereoscopic images, the processing of stereoscopic images would reduce their quality inevitably in each link of image processing system. Stereoscopic image quality assessment can not only affect the subsequent processing of stereoscopic images but also evaluate and reflect the property of multitmedia system. Therefore, the research is very important in auto stereoscopic image system. Based on the current research of the HVS’s perceptual characteristics, this paper proposes two kinds of stereoscopic images assessment, which are based on the characteristics of HVS depth perception. And the main contents of this paper are briefly described as follows:(1) Based on the research of the perceptual characteristics of HVS, the stereoscopic images can be devided into different depth planes according to the depth map. HVS has different perceptual sensitivity to the planes, which have distinct depth values in the depth map. And the planes with different sensitivity play different roles in the HVS’s ultimate assessment of stereoscopic image quality. This makes those different planes would have different weight coefficients. The distortion on every depth planes is measured by the difference of the singular values between the original and distorted images, through singular value decomposition. The global distortion score is finally calculated by coalescing the quality scores of different depth planes with weighting mechanism.(2) Based on the perceptual characteristcs and the 2D image quality assessment, a model with quaternion expression for assessing the quality of stereoscopic images is established. This method aims at color images through representing the color image pixel with a quaternion, which accesdes the chromincance and lumincance information to the calculation of distortion simultaneously. Furthermore, the structural distortion and energy distortion are separated and calculated respectively. The statistic of structural distortion not only includes the changes of spatial location but also with the spatial rotation joined, which makes the algorithm can measure the distortion of the color images comprehensively.After verified the proposed algorithms in the public database, the two methods both have excellent experiment results. They have important significance for both application and research in stereoscopic image system, and offer guidance for the subquent development of stereoscopic image quality assessment.