Structured Video Coding

In order to transmit a video signal over a limited bandwidth channel many techniques have been developed which can reduce the quantity of data required to represent moving image sequences. Almost all current coders use techniques based on reducing the statistical redundancy of the signal and psychophysically less significant details. Such coders are relatively successful for coding signals at bandwidths of 1 Megabit/second and higher but below this level purely signal based techniques become much less robust. This thesis will attempt to use an analysis of the structure of the image, the components from which it is composed, in order to gain greater image compression. The input material is examined and divided into elements representing the background, the actors and any other significant independent objects. The background image is stored as a model at the receiver and animated according to estimates of the motion in the original scene. Actors and other foreground objects are segmented out of the scene, compressed and transmitted using conventional signal based methods. This technique typically leads to reductions in bandwidth of greater than 70%. In addition, enabling the computer to distinguish the components of the picture allows for forms of post-shoot modifications of the picture not possible with conventionally represented signals. Thesis Supervisor: Andrew Lippman Title: Associate Director, MIT Media Laboratory The work reported herein was supported by a contract from the Movies of the Future consortium, including Apple Computer, Inc., AT&T, Eastman Kodak Co., Viacom International Inc., and Warner Brothers Inc.