Shell representation and compression conscious manipulation for three-dimensional graphical datasets

Abstract

Although there have been a large number of standards for two-dimensional images, there has not been as extensive a study into representing three-dimensional datasets. Presented here is a test from an extended system to cater for initially autostereo sequence datasets, although still suitable for all types of three-dimensional datasets. Extensions to do three-dimensional manipulation operations are defined that are both computationally efficient when applied to the coded form and are compression conscious in that they maintain compression performance. Autostereoscopic displays have been developed which create a three-dimensional image without the need for special glasses or headsets. They work by displaying multiple images each viewable over a very narrow field of view. This has the advantage of being able to look around objects. Increasing the number of images per view increases the bandwidth required. To aid stereo disparity in the human visual system the important features to retain in stereo sets are potentially different from standard two-dimensional images. Edge information and highlights have been shown to be of paramount importance for the human visual system to accurately perceive depth. The theme of coding three-dimensional regions as separate entities allows for a potential high compression ratio and subsequent transmission rate speed-up. A controlled merging scheme of regions is also presented which, being designed to keep all highlights and avoid defocusing of the image, gives a corresponding compression ratio increase while retaining the majority of the stereo cues.