S. Lepsøy, P. Carlini, G. E. Øien. On fractal compression and vector quantization. In Conf. Proc. NATO ASI Fractal Image Encoding and Analysis, Trondheim, July 1995, Y. Fisher (ed.), Berlin Heidelberg, 1998.
Fractal image compression can be seen as a type of vector quantization with a codebook that is extracted from the image being coded. In this chapter we study the properties of codebook populated in this way and make comparisons with fixed codebooks that are trained for a set of representative images. There are two major results of the experiments presented here. Firstly, the image being coded is not necessarily the best codebook-generating image; the codebook can be as good or better when it is extracted from another image. Secondly, we have compared a fractal coder to a very similar product code vector quantizer with the original of the codebooks as the most significant difference. In terms of PSNR of decoded images, the product code VQ has been superior to the fractal coder in all 108 tests, over six test images, nine codebook sizes and two block sizes. When a coding technique is made adaptive, the adaptivity should improve, not deteriorate the results obtainable by the technique. The fractal coder is an image-adaptive vector quantizer, and should as such offer an advantage over its non-adaptive counterpart. Thus, the main message is not that there exists a certain VQ that is slightly better than the fractal coder tested here, but that fractal coding may be the wrong way of codebook adaption.
@InProceedings{LeCaOi98,
Author = {Lepsøy, S. and Carlini, P. and Øien, G. E.},
Title = {On fractal compression and vector quantization},
BookTitle = {Conf. Proc. NATO ASI {Fractal Image Encoding and Analysis}, Trondheim, July 1995},
editor = {Fisher, Y.},
Publisher = {Springer-Verlag},
Address = {Berlin Heidelberg},
Year = {1998}
}