Set -partitioned image coding

Hierarchical subband coding has been very successful over recent years in the compression of still and moving images. In particular, set-partitioned hierarchical subband coding schemes have been found to yield excellent results, yet have low complexity—a phenomenon which has paved the way for fast and efficient coders. In this thesis, we analyze popular set-partitioned image coding schemes, namely SPIHT and AGP, to understand the reasons for their effectiveness. We develop information-theoretic rate equations and compare the results with the first order entropy of the stochastic models used for image analysis. It is known that the effectiveness of a coding scheme can be gauged truly if the entropy rate of the source is known. This is a nearly impossible task for natural image sources. An attempt is made to estimate the entropy rate of such sources by making some simple and reasonable assumptions. We define a new concept in image coding by developing the first truly embedded block-based image coder. This new coder, called the Set-Partitioned Embedded bloCK coder (SPECK), is perfectly embedded, has low-complexity yielding very high coding speeds, and a PSNR performance which is among the best in the world of low-complexity coders. We perform an analysis of this coding scheme, and also develop a modified version of it—the Low-Memory SPECK—that codes blocks of image independently of each other, thereby reducing memory consumption from the order of image size to block size. The JPEG2000 Still Image Compression Standard—the next-generation still image compression standard currently under development and open to proposals—requires the forthcoming standard to conform to certain requirements and profiles. The Low-Memory SPECK coder meets most of the JPEG2000 requirements very well and has been proposed as a serious candidate for adoption as the next-generation still image compression standard.