A new perceptual quality metric for video transrating for mobile devices

In video streaming over heterogeneous networks to mobile devices, severe bandwidth mismatch creates significant challenges in maintaining quality of experience for end users. The reduction of the video bit rate to address the problem of bandwidth mismatch is usually achieved by sacrificing the video quality. Most existing rate control approaches for transrating are based on the Mean Square Error (MSE) metric, which often leads to considerable degradation of the quality of user experience. This is due to the fact that the MSE seldom corresponds well to perceptual quality. This research explores the possibility of adapting the bit rate of video streams to suit varying network conditions while maintaining as high as possible the perceptual quality. In this paper we present a novel perceptual quality metric, Average Relative Perceptual Distortion (ARPD), based on the just noticeable distortion (JND) theory and developed a model to weight the perceptual importance of video components. We further applied this model into the development of a MPEG transrater that can requantize video components with proper weights in order to optimize the overall video perceptual quality. The results obtained from subjective tests conducted on a set of different video streams show the proposed scheme is able to improve the perceptual quality of video transrating, as well as maintain the consistency between the proposed metric ARPD and the subjective scores.