Autocorrelation analysis for reduced-reference mobile video quality assessment

Due to the growing use of mobile video services, monitoring of perceived video quality in live systems becomes of high interest. For this purpose, it is required to capture video quality consistent with the principles behind the operation of the human visual system and to accommodate some suitable quality reference to the transmission of the mobile video service. This can be achieved by extracting structural information in terms of a finite number of spatial features from a video sequence and utilize these features as foundation of a reduced-reference design of a perceptual video quality metric. In this paper, we examine the dynamics of such spatial features with respect to their progression over time by applying an autocorrelation analysis that reveals the coherence time over which the individual features can be considered as constant. As prerequisite of the analysis, we have conducted subjective tests to categorize a set of standard videos into the classes of slow, median, and fast motion. The coherence times deduced from the autocorrelation analysis show that the dynamic behavior of certain spatial features is independent from the video content while statistics of other spatial features are dependent on the video categories. These findings allow us to reduce not only the computational load for capturing dynamics of spatial features in video sequences at the mobile device but also the amount of information to be carried as reduced-reference to the receiver.