Visual Saliency Detection And Its Application In Video Quality Assessment

Visual significance is an inherent property of the human visual system.Significant areas represent the position of eye’s attention, is a key factor to evaluate the perceptual video quality. According to HVS characteristics to detect significant area, and apply it in the video quality assessment can greatly improve the video quality. The perceptual quality of the video is also closely related to the distortion.How to apply the significant and distortion areas to the video quality assessment is also the research content of this article. The human eyes are the ultimate recipient of the video, so the subjective video quality assessment is the most direct and effective method. But due to the subjective evaluation is time-consuming and laborious,resulting that it can’t be applied to real-time processing. Therefore, we must find an objective evaluation method to meet the human visual characteristics.This paper investigates the existing algorithms of video significant detection and video quality assessment, analyses the application and shortage of the visual saliency detection, found that the video time-domain correlation and the persistence of vision are the key factors to the video significant. In this paper, combining with the space-time domain correlation and the persistence of vision, through establishing time-domain sliding window to detect the significant region in the three dimensional frequency domain. Finally we can get the saliency area through connectivity analysis and adaptive threshold judgment. Frequency-domain algorithm’s speed is very fast,thus ensuring the effectiveness of the proposed algorithm. Experiment results show that the algorithm improves the accuracy and also reduces the computational complexity.Thesis also analyzed the saliency area, found that if given saliency region a large weight can get better video quality. So we must turn this area into the weight coefficient in order to adjust the video quality assessment. In this paper, we use the Gaussian surface to model the saliency area in order to get the weight coefficient.The weight will be applied to the video quality assessment to make subjective and objective evaluation tends to be more consistent.Based on the visual saliency detection, we summarize and further extend the application of the visual significant, and make the significant area apply to the video quality assessment. In this paper, the 3D-SSIM algorithm is proposed based on the motion characteristics and the temporal correlation of video. The video is seen as a three-dimensional image, and the video is divided into 3D blocks on which to calculate the 3D-SSIM score. Later the significant and distortion weights are integrated to the 3D-SSIM algorithm, so to increase the evaluation score. Experiment results show that in the case of low complexity, the algorithm makes the objective evaluation closer to the subjective feeling of human’s eyes. These results proved the feasibility of the algorithm, but also further verified the significant area’s promotion effect on the video quality assessment.