HMM Based Signature Identification System Robust to Changes of Signatures with Time

This paper describes the signature identification using long-term signature database. In signature identification, changes of signatures with time give serious influence on the identification rate and requests the users whose signature is changeable with time write additional signatures for updating reference. We proposed signature identification using HMM and performed experiments using large signature database obtained from 170 persons in 50 days. The result clarified the relationship among identification rate, training depth, and robustness against changes of signatures with time. Although 90% of identification is obtained with 4 days training, it is also found that the identification rate degrades as time passes. With 20 days training, however, the improvement after 20 days training becomes more than 7% and 90% of identification rate is obtained.