结合人眼对比度敏感视觉特性的视频质量客观评价

结合人眼对亮度、色度、对比度以及运动目标的感知特性,提出了一种基于人眼对视频内容感知的视频质量客观评价方法。该方法将视频分为空域和时域信息分别描述,并利用人眼感知特性,从视频的亮度、色度、对比度以及目标运动4个方面提取特征,计算其强度。然后以人眼对比度敏感值作为强度的权重因子求和,构建人眼感知视频内容模型。最后,分别以此模型模拟人眼感知源视频和失真后的视频,计算每对应单元的所有像素之间和运动矢量之间的强度差;以强度差作为视频质量评价的分数,构建视频质量客观评价模型。采用LIVE数据库中的6个源视频和48个测试视频进行了质量评价实验,并与视频质量专家组(VQEG)推荐的5个较好的视频质量客观评价模型进行了对比分析。结果表明:提出模型的视频质量评价结果与主观评价结果之间的线性相关性系数达到0.8705,显示了较好的一致性,评价效果优于5个典型的模型。

Video quality objective assessment combined contrast sensitivity characteristics of human visual system

In combination of the perceiving characteristics of human eyes for brightness, chroma, contrast and moving targets, an objective assessment method of video quality based on contrast sensitivity characteristics of a human visual system was proposed. In the method, the video was divided into spatial and time domains to be described. The features of image were extracted from four aspects, brightness, chroma, contrast, and target motion based on the perceiving characteristics of human eyes and their intensities were computed. Then, the contrast sensitivity values of human eyes were used as the weight factors of the intensity to sum and to construct the model of human eye perception content of the video. Finally, original and distorted videos respectively perceived by imitating eyes with this model, and the intensity differences of the pixels and the motion vectors between arbitrary corresponding units of two videos were computed. By taking the intensity differences as the scores of video quality objective evaluation, the objective evaluation model for video quality was constructed by them. The experiments were carried out with 6 source videos and 48 test videos proposed by LIVE database, and the 5 classic video quality evaluation models recommended by the Video Quality Expert Group(VQEG) were compared with the proposed model. The results show that the linear correlation coefficient between video quality evaluated by the proposed model and the subjective evaluation results reaches 0.8705. They have good consistency, and evaluation effects are better than those of other 5 classical models.