基于SOFM/HMM模型的非特定人手语识别系统

手语识别是通过计算机提供一种有效而准确的机制将手语翻译成文本或语音。目前最新发展水平的手语识别系统在实际应用中应该解决非特定人手语识别问题。该文在分析非特定人手语识别特点－数据多且差异大、模型训练难收敛、对不同人数据的特征提取需求更迫切－的基础上，提出了SOFM／HMM模型，将自组织特征映射（SOFM）很强的特征提取功能和隐马可夫模型（HMM）良好的处理时间序列属性结合在一个新颖的框架下，并把该模型应用到非特定人中国手语识别中，实验结果表明，SOFM／HMM模型手语识别率比传统的HMM模型提高近5％。

A Signer-Independent Sign Language Recognition System Based on SOFM/HMM

Sign language recognition has emerged as one of the most important research areas in the field of human computer interaction. The aim of sign language recognition is to provide an efficient and accurate mechanism to transcribe sign language into text or speech so that communication between deaf and hearing society becomes more convenient. State of the art sign language recognition should be able to solve the signer independent problem for practical applications. This paper analyzes the features of signer independent sign language: (1) the convergence difficulty caused by mass data and noticeable distinctions between different people data. (2) the urgent need to extract common features from different people data. Aiming at these features, the SOFM/HMM model presented in this paper combines the powerful feature extraction performances of self organizing feature maps (SOFM) with excellent temporal processing properties of hidden Markov models (HMM) within a novel scheme. Each SOFM eigenveter centroid is regarded as one of the components in the state of HMM which construct the state probability density function in terms of the weighted sum. The model parameters can be re estimated through the Expectation Maximization (EM) algorithm. When the proposed model is applied to signer independent Chinese Sign Language (CSL) recognition with a vocabulary of 208 signs, 95.3% recognition rate is obtained in the registered test (Reg.) and 88.2% in the unregistered test (UnReg.). Meanwhile, results from the conventional HMM system are provided as comparison. Experimental results show the SOFM/HMM system increases the recognition accuracy by 5% than conventional HMM one.