Stereoscopic Image Quality Assessment

3D imaging is a wide research area driven by the entertainment industry andthe science applications. Stereoscopic technologies have burgeoned which led to anincreasing interest in various3D applications. The3D image delivery chainincludes the stages of capture, encoding, transmission, possible post-processing atthe receiver side and then display. Any of these stages may cause degradation of3Dvisual quality or errors introduced at a certain step of the production process maypropagate through the chain. Therefore,3D image quality assessment is a keyfactor in the design and optimization parameter of3D image processing systems.This paper will introduce objective quality assessment method of stereo image.According to the experimental analysis of physiology in human visualcharacteristics, this paper puts forward the methods by combining JND with thecharacteristics of visual attention to evaluate the quality of the image. First of all,JND model can measure the visible threshold of the HVS. According to thethreshold value, this paper removes the visual redundancy of image so that theevaluation result is more accurate. And in an image, only some of the area canattract our eyes. Therefore, these areas could be extracted according to saliencytesting. At the same time, the area of poor quality can also cause our attention, wealso find these areas. Second, the image will be divided into four areas: bothsaliency area and inferior area, only saliency area, only inferior area, neithersaliency area nor inferior area. Again, the four areas endow with the differentweighting to compose the visual importance model (VS). Finally, the VS model isfused to SSIM. Stereo image quality assessment methods are very scarce and thestereo image database is not available. So this paper validates the performance ofthe VS model on LIVE database. The experimental results show that this method isgenerally competitive with current quality assessment metrics of image.The basic principle of3D image technology is to observe the two views usingthe left and right eye with parallax, and the two views are associated due to thedisparity of observation. So there is a realistic three-dimensional world in thehuman brain. In this paper, the3D image quality assessment is divided into twoparts: quality evaluation and stereo sense evaluation. Quality evaluation continue touse signal image quality evaluation method, respectively, quality evaluation resultis defined the arithmetric mean of the left and right image assessed values. Thedifferences about the left and right image caused stereo sense, and absolute disparity image is the most direct reflect. Therefore, the stereo sense ofstereoscopic image is evaluated using the absolute disparity image. Through theexperiment, stereo image objective evaluation result is accordance with subjectiveevaluation results.