关键词: 小波包, 高斯模型, 隐马尔可夫模型, 状态转移, 聚类

Abstract: The paper proposed a way of off-line handwritten signature recognition based on wavelet packet and Hidden Markov Model (HMM). Wavelet packet was used to extract the features for the whole normalized signature image; the distribution of the wavelet packet coefficients could be approximated by mixture Gaussian model, and the state transfer model of HMM was adopted to describe the relevancy and dependency of each channel in the mixture Gaussian model. The data preprocessing is simple, and the feature extraction is complete and reversible. This method avoided complex segmentation and illuminated the decomposed statistical characteristics of the signature image. The experimental results show that the algorithm has better anti-noise ability, robustness and the recognition rate is higher.

Key words: wavelet packet, Gaussian model, Hidden Markov Model (HMM), state transfer, clustering