Proceedings of the International Workshop on Applications of Neural Networks to Telecommunications

Vector Quantization Using Tree-Structured Self-Organizing Feature Maps

Tzi-Dar Chiueh, Tser-Tzi Tang, and Lian-Gee Chen

Department of Electrical Engineering National Taiwan University Taipei, Taiwan [e-mail]

Abstract

In this paper, we propose a modified SOFM model with a binary-tree cell structure. During and after training, the centroids of the neurons in the model always constitute a binary tree in the higher-dimensional input space. This model is used to design codebooks for vector quantization coding of images. Simulation results show that the acquired codebook is more diversified and has a tree structure. With these two characteristics, it not only produces better-quality images but also expedites the encoding (quantization) ...