Research On Stereoscopic Image Quality Objective Assessment Methods Combined With Binocular Visual Characteristics

Stereoscopic video, which can provide us more realistic visual effects, has been widely known.Because the space we live in is three-dimensional, stereoscopic image\video can bring peopleimmersive feel. However, the acquisition, encoding, transmission and many other aspects of thethree-dimensional video coding system induce the decreasing of image quality of receiving end.Thus, stereoscopic image\video assessment has become a research focus. This paper utilizes thebinocular visual characteristic and research stereoscopic image quality assessment methods ofthree-dimensional video coding system. The main contents are as follows:Firstly, by simulating the change of the distance between cameras, this paper firstly proposesthe three-dimensional disparity spatial image and establishes a three-dimensional structuralsimilarity evaluation model. Then we extract quality of regions of serious distortion and use it asthe quality of the whole image. Next, considering the impact of binocular just-noticeable-differenceand local disparity, this model has been improved further and obtains better performance.Secondly, considering the characteristics of binocular vision, this paper proposes twostereoscopic image quality assessment models. Model A: the image losses useful information andincreases redundant information when distorted. This paper calculates the detail information ofimage by wavelet transform to gain restore image and disturbed image. For different characteristicsof image using different assessment methods, this paper utilizes binocular fusion and inhibitioncharacteristics combining with quality of left-right views to get the quality score. The perceptionof human eyes is different for different regions in stereoscopic image, this paper dividesstereoscopic image into single and binocular regions based on whether the contents of the left-rightimages are matched with each other. Finally, it uses the Gabor filter to extract perceptioninformation from different regions and integrates this information to obtain the final image qualityobjective value. In order to prove the performance of the model, this paper conducts experimentson symmetrical stereoscopic image database and it uses two different databases to getcross-validation.Finally, after the research of stereoscopic image content quality assessment, we propose twomethods of stereoscopic image assessment from angle of depth perception.1) This paper calculatesthe absolute difference image of stereoscopic image and then conducts singular valuedecomposition of them to gain the singular vectors to rebuild the original and distorted images. The phase difference characteristics of rebuilt images are used to establish the objective evaluationmodel.2) Binocular energy information of stereoscopic images is calculated by Gabor filter. Thephase difference of binocular energy information and gradient information are applied to computeparallax offset of stereoscopic images. The second approach, redefined parallax shift using thephase difference of the energy image, achieves good performance.