Signature Identification Using Dynamic and HMM Features and KNN Classifier

Dynamic feature-based signature recognition systems obtain more identification accuracy rate than static feature-based signature recognition systems. On the other hand, because of point to point comparing of captured signals, local features produce less error than general features. In this paper, a novel signature identification algorithm based on dynamic and Hidden Markov Models (HMM) features and K-Nearest Neighbor (KNN) classifier is proposed. In this algorithm, special point dynamic features of handwritten signatures are selected and normalized and then used as signature features. In the used HMM, different number of states in transition matrix for each signature are calculated and used as feature vector. This strategy not only reduces complexity computational but also is extracted features based on general characteristics of signature. The proposed algorithm is evaluated on the standard SVC2004 database. The obtained results are reported based on various values of parameters of the proposed algorithm and compared with presented approaches in the literature.