Computer Vision Techniques for Hidden Conditional Random Field-Based Mandarin Phonetic Symbols I Recognition

Abstract

This paper presents a handwritten recognition method using camera as human-computer interaction device (HCI) for Mandarin Phonetic Symbols I (MPS1). The method is based on a hidden conditional random field (HCRF) model, which is an extension of the conditional random field (CRF) framework that incorporates hidden variables. The main advantage of the proposed method is that it avoids limitations of the traditional hidden Markov model (HMM)-based methods. This work built an HCRF for each symbol of MPS1 and used twelve-dimensional features. The features in the proposed system include the stroke length ratio feature, the horizontal stroke feature, the vertical stroke feature, the stroke-based loci features, and the stroke curvature feature. The recognition rate achieved 94.05% on 1532 handwritten word samples covering 37 symbols.