Block-based attractor coding: potential and comparison to vector quantization

Abstract

The paper presents a simple fractal or attractor coder model that has a very fast decoding algorithm and lends itself to comparisons with vector quantization (VQ) of the mean-gain-shape (MGSVQ) type. In fractal theory the transmission of the codebook is somewhat concealed. In our simple model the codebook is explicitly transmitted although with a double role. The main difference between MGSVQ and the fractal coder is that the codebook in MSGV is as statistically optimized from a set of training data whereas it is derived directly from the image to be coded for the fractal coder, and therefore can be viewed as adaptive. Experimental comparisons are given.<>