Human Visual System Based Stereoscopic Image Quality Assessment

With the rapid development of multimedia technology, stereoscopic imagingtechnology came into being. Compared with the monoscopic images, stereoscopicimages can provide additional depth information which can provide the feeling ofbeing personally on the scene, so in the entertainment, military, industrial and variousother fields begin to play an important role. With the large-scale development ofstereoscopic images, many applications need the ability to quickly and efficientlycomplete the stereoscopic image quality evaluation for further applications, but thesubjective quality assessment is difficult to meet the requirements in terms ofefficiency. Therefore, this paper proposes two objective stereoscopic image qualityevaluation methods which explain the meaning of stereoscopic image quality fromtwo different perspectives.Different from the monoscopic image quality, the stereoscopic image qualityhas its unique evaluation index: stereoscopic quality. It emphasizes the quality ofstereo vision, rather than the distortion between the viewed stereo and the capturedstereo. Even if the stereoscopic images are not degraded, they may produce differentperceptual quality scores also. Therefore, we propose a perceptual quality evaluationalgorithm, which combines some characteristics of the human visual system. Firstobtain the disparity map, then adjust the disparity map through the weights in theedge map and the saliency map. After that the Minkowski pooling is used to integratethe weighted disparity map. Finally, the multiscale strategy is applied to compute thefinal score. The EPFL database is utilized to validate the proposed metric.Experiment shows that the objective score obtained by the proposed metric and thesubjective scores have a high degree of consistency and monotonicity. It is provedthat the algorithm to evaluate the perceptual quality of stereoscopic images in thispaper is very effective.Like the monoscopic images, there are always all kinds of distortions duringcapturing, encoding, transmission and decoding process in the stereoscopic images.The structural similarity index for the evaluations of distortions in the monoscopicimages proposed by Zhou Wang achieves good experimental results. However, dueto the unique characteristics of the stereoscopic images, applying the structuralsimilarity index of monoscopic image quality evaluation method directly to the stereoscopic images can not achieve the most desirable results. In this paper, somecharacteristics of the human visual system are added into the structural similarityquality evaluation algorithm. First the stereoscopic images should be transf ormedfrom the RGB space to the CIELAB space to remove the chrominance component.Then use the wavelet transform to eliminate the influences caused by distortions inthe horizontal directions, and use the precise depth information to reduce the impactof the distortion in the non-major regions. Finally the experiment on the LIVE3DIQA database shows that the proposed distortion quality evaluation algorithm hascertain advantages compared with some other algorithms.