Continuous HMM applied to quantization of on-line Korean character spaces

Abstract

This paper demonstrates the effectiveness of a continuous HMM (CHMM) for quantizing on-line Korean character spaces. Vector quantization (VQ) error is a major factor that affects the performance of pattern recognition. Accordingly, a CHMM is used to classify on-line Korean character space into clusters where each cluster is represented by a CHMM state Gaussian function. The experimental results show that the proposed CHMM vector quantization decreases the quantization distortion in the VQ stage compared to other methods and thereby improves the performance of a discrete HMM-based recognition system.