Characteristics of disparity-compensated stereo image pair residuals

摘要

An emerging feature of multimedia and tele-presence systems is stereo imagery. Stereo images provide an enhanced sense of presence, and have been found to be operationally useful in tasks requiring remote manipulation or judgment of spatial relationships. A conventional stereo system with a single left–right pair needs twice the raw data as a monoscopic imaging system. A multi-view stereo system may require hundreds of times as much. As a result there has been increasing attention given to image compression methods specialized to stereo pairs. Most of this work has focused on disparity-compensated residual coding where one view is used to predict another, and the difference is coded. Much of this work has concentrated on improving the disparity-compensation process and codes the residual image similarly to a monoscopic image. As this paper will show, the disparity compensated residual image differs in important ways from a conventional monoscopic image. The residual image has different correlation characteristics, and is composed primarily of strong vertical-direction edge components surrounded by large areas of near zero intensity. This paper demonstrates these properties, both theoretically and experimentally, and outlines how they might be exploited in designing compression algorithms. Other papers demonstrate that greater than 2 dB improvement is possible when exploiting such characteristics for stereo image compression.