ABSTRACT
The objective of image compression may be summarized as that of finding good approximations to the original image that can be compactly represented. The key questions are: (1) how can we choose “good” approximations? and (2) how can we represent the chosen approximations with as
few bits as possible? In recent years, scalable compression has proven to be of particular interest. A scalable compressed bit-stream contains multiple embedded approximations of the same image, with the property that the bits required to represent a coarse approximation form a subset of the bits required to represent the next finer approximation. Scalable compressed bit-streams support successive refinement, progressive transmission, accurate rate control, error resilience, and many other desirable functionalities. Scalability gives rise to a third important question: (3) how can we choose a sequence of good approximations, whose representations are embedded within each other?
