Optimization of DHMM Based on Chaotic Migration-Based GA for Chinese Signature Verification
Zhenhua Wu · 2007
In this paper, Genetic Algorithm (GA) is used to train the parameters of Discrete Hidden Markov Model (DHMM).To overcome the premature convergence in GA, a chaotic migration strategy is introduced to the pseudo parallel genetic algorithm to increase the diversity of population.Because the GA's evolution speed is very slow, the Baum-Welch is applied to the GA.A floating matrix encoding mechanism is used for reflecting internal relations consisted in parameters of DHMM.This encoding method reduces the searching range of solutions space and increases the searching efficiency further.By using GA, the number of states can be adjusted dynamically.At last, the proposed method is used for signature verification.The promising experiment result indicates that the chaotic migration-based GA can optimize DHMM effectively.