Coding of Segmented Image Sequences
Ujjaval Y. Desai, Ujjaval Y. Desai · 1994
Standard image sequence coding methods, such as MPEG, use block-based motion compensation and transform techniques to reduce temporal and spatial redundancies. Higher compression, however, can be achieved by object-based image sequence representation. In an object-based representation scheme, the image sequence is decomposed into objects on the basis of motion, texture, structure, etc. By efficiently encoding these arbitrarily shaped objects, significantly lower bit rates can be achieved. We propose an object coding method which separately encodes the intensity and shape of objects. A compact description of the shape is obtained by using a contour encoding algorithm, while efficient transform coding techniques are used to compress the intensity. We describe three such techniques: smooth filling, cosine filling, and region dependent transform coding. We compare these methods with the KLT. Thesis Supervisor: E. H. Adelson Title: Associate Professor, M.I.T. Media Lab Acknowledgements Fi...