Generating video object planes for MPEG-4 using vector quantization

The popularity of digital television and video conferencing has necessitated the need for good video compression. The Moving Picture Experts Group (MPEG) has introduced many compression standards including MPEG-4, which uses an object-based model for coding. The extraction of individual objects from a frame is one of the primary functions of the MPEG-4 encoder.; Since MPEG-4 is a decoding standard, encoding operations like object extraction have not been standardized. The watershed algorithm is a popular technique used for object extraction. Though the watershed algorithm has many advantages, it suffers from the problem of oversegmentation. This thesis uses the pattern-matching property of vector quantization to perform edge detection. A simple strategy to generate the complete object using the edge information has been demonstrated, and the results of these have been compared with the watershed algorithm. The encouraging results confirm that vector quantization can find applications in the MPEG-4 encoder.