Knowledge-based system for coding head and shoulder images

Several implementation issues of the primitive-based image codec are discussed, namely, the structure of the database, coding the primitives when no match is found in the database, buffer control, and accumulation of error. A hierarchical database, constructed during the coder´s operation, is used to avoid difficulties of constructing an offline database. Singular-value decomposition of primitives is used in a way similar to that of transform coding to construct a compact database and to code the primitives when no match from the database is found. Accumulation of error is prevented by transmitting with each frame the pixels from three lines that are evenly spaced in the image raster and moved up each frame time. The achievable compression ratio is also analyzed