We present a novel technique for efficient visual communication

where a compact representation is abstracted from arbitrary image sequences automatically.We aim at improving the entire visual communication process

including transmission of video data and perception of visual signal by human eyes.Therefore

guided by some related theory in Psychology

our work increases both the compressibility and comprehensibility of images.This goal is achieved by two steps.One step is an edge extraction process for preserving object boundaries which are believed to be the most sensitive features to human vision

and the other step is a non-linear diffusion process for reducing extraneous details.Both are designed in a spatiotemporal manner to guarantee the temporal coherence of resulting animations

while real-time processing speed is maintained by facilitating parallel computation on a GPU.We additionally build a complete visual communication system with the proposed algorithm as its core to demonstrate the practicality of our technique.Experimental statistics collected on the system indicates that transmission bandwidth is obviously saved while perceptibility of image sequences is improved.