基于时域处理的视频质量评价方法

Video quality assessment based on temporal processing

目的利用人眼视觉系统时域处理机制开发视频质量评价算法。方法利用人眼视觉系统的饱和与不对称感应原理,对连续帧质量变化进行限制,然后对单帧质量进行时域处理产生视频质量指标。结果提出84个视频质量评价指标(包括12种单帧估计和7种时域处理方法)。结论单帧质量评估方式的优劣对于视频质量评价算法的整体性能至关重要,所有性能良好的视频质量评价算法均基于与主观质量高度吻合的图像质量评价方法;表现最好的视频质量评价模型分别是利用百分位数、明可夫斯基求和,加权明可夫斯基求和与本地L帧的最大值作为时域处理方法。

Aim To develop video quality assessment algorithm based on spatial temporal processing mechanism of the human vision system. Methods Apply the saturation and unsymmetry properties of the human vision system to limit the single frame quality, which is then temporally processed to generate video quality indicators. Results 84 video quality indicators are proposed using 12 single frame quality estimation and 7 temporal processing methods. Conclusion Single frame quality estimation methods are key for the performance of video quality assessment algorithms, all the video quality assessment algorithms with good performance are developed based on image quality as sessment algorithms agreeing well with subjective ratings; the best video quality assessment algorithms employ per- centile, Minkowski sum, weighted Minkowski sum, the mixamum of L local frame quality as the temporal processing methods.