Multi-class maximum entropy coder
R.D. Dony, S. Haykin · 2002
The optimal linear block transform for coding images is known to be the Karhunen-Loeve transform (KLT). However, the assumption of stationarity in the optimality condition is far from valid for images. Images are composed of regions whose local statistics may vary widely across an image. The authors propose a new transform coding method which optimally adapts to such local differences based on an information-theoretic criterion. The new system consists of a number of modules corresponding to different classes of the input data. Each module consists of a single-component, linear transformation, whose basis vector is calculated during an initial training period. The appropriate class for a given input vector is determined by the optimal maximum entropy classifier. The performance of the resulting adaptive network is shown to be superior to that of the optimal nonadaptive linear transformation, both in terms of rate-distortion and computational complexity.