Abstract
Video compression algorithms reduce the bit rate needed to transmit video signals. At bit rates relevant to practical transmission scenarios, the compressed video signal differs from the uncompressed one. The compressed signal typically is of lower quality and the degradation relative to the original represents a perceived distortion. Here we introduce an evoked potential method that quickly and objectively measures differences between video signals that are correlated with perceived distortion. A video sequence was subjected to increasing levels of compression through an intra coding method (High-Efficiency Video Coding standard H.265/MPEG-HEVC). To generate an evoked response related to perceptual differences, we alternated between uncompressed and compressed video sequences at 3 Hz. A given level of compression was presented for 1 sec and the degree of compression was increased in equal-sized steps every second, yielding a 16 sec trial. The participants (n=8) pressed a button when they first detected compression modulation. The evoked potential increased monotonically at the 3 Hz compression alternation rate and its harmonics. Thresholds estimated from these functions were well correlated with the behavioral threshold. The scalp topography of the first and second harmonics suggested the presence of at least two mechanisms sensitive to the presented differences. We modeled the generation of the first and second harmonics using a simple image-difference model—by taking the sum of the absolute differences between each pair of frames. This model produced two image-based signals analogous to the evoked potential harmonics. The first-harmonic signal reflected the sustained response to the level of compression, which increased over the trial. A second-harmonic signal reflected the transient differences between compressed and uncompressed frames, which also increased as compression-level increased. Evoked responses thus provide an efficient and accurate measure of two distinct neural correlates of perceptual differences.
Meeting abstract presented at VSS 2014