On-Line Handwriting Signature Verification Based on Parameters Optimization of HMM
Liting Zhang, Jianbin Zheng, Enqi Zhan · 2010
Hidden Markov model (HMM),as a representative of statistical model method,can well describe the problem of pattern recognition in time series process,such as handwriting signature verification or speaker identification.But it has universal low verification rates,a long time of training and other disadvantage when used in on-line handwriting verification. Through experiments,we find that the topological structure of HMM,the state numbers and the Gaussian components have an important influence on verification rate,and we also propose a parameters optimization method of HMM.The result achieves a false rejection rate (FRR) of 4.10% and a false acceptance rate (FAR) of 2.82%.