In this chapter, we introduce three nonstandard image coding techniques: vector quantization (VQ) (Nasrabadi and King, 1988), fractal coding (Barnsley and Hurd, 1993; Fisher, 1994; Jacquin, 1993), and model-based coding (Li et al., 1994). INTRODUCTION The VQ, fractal coding, and model-based coding techniques have not yet been adopted as an image coding standard. However, due to their unique features these techniques may find some special applications. Vector quantization is an effective technique for performing data compression. Theoretically, vector quantization is always better than scalar quantization because it fully exploits the correlation between components within the vector. The optimal coding performance will. | 9 Nonstandard Image Coding In this chapter we introduce three nonstandard image coding techniques vector quantization VQ Nasrabadi and King 1988 fractal coding Barnsley and Hurd 1993 Fisher 1994 Jacquin 1993 and model-based coding Li et al. 1994 . INTRODUCTION The VQ fractal coding and model-based coding techniques have not yet been adopted as an image coding standard. However due to their unique features these techniques may find some special applications. Vector quantization is an effective technique for performing data compression. Theoretically vector quantization is always better than scalar quantization because it fully exploits the correlation between components within the vector. The optimal coding performance will be obtained when the dimension of the vector approaches infinity and then the correlation between all components is exploited for compression. Another very attractive feature of image vector quantization is that its decoding procedure is very simple since it only consists of table look-ups. However there are two major problems with image VQ techniques. The first is that the complexity of vector quantization exponentially increases with the increasing dimensionality of vectors. Therefore for vector quantization it is important to solve the problem of how to design a practical coding system which can provide a reasonable performance under a given complexity constraint. The second major problem of image VQ is the need for a codebook which causes several problems in practical application such as generating a universal codebook for a large number of images scaling the codebook to fit the bit rate requirement and so on. Recently the lattice VQ schemes have been proposed to address these problems Li 1997 . Fractal theory has a long history. Fractal-based techniques have been used in several areas of digital image processing such as image segmentation image synthesis and computer graphics but only in recent years have they been extended to the .
