Signature verification system based on multiple classifiers and multi fusion decision approach

Abstract

With an increase in identity fraud and the emphasis on security, there is growing and urgent need to verify human identify efficiently. Signature and the handwriting verification application are used in many fields such as banking, public sectors. Documents and cheques verification system has triggered a real need for reliable, accurate and robust system. This work adopts different classification techniques between the local features based and the global features based of the signature system in addition to different fusion techniques between the outputs of the different classifiers and global features based to improve error rate of behavioral system. Main goal is to develop more accurate and robust signature verification system than the previous developed system with False Rejection Rate (FRR) equals to 5.3 and False Acceptance Rate (FAR) equals to 0. To achieve this goal, first multiple classification techniques are applied to the signature verification system which are artificial neural network, support vector machine and Pearson correlation and then these techniques are fused by applying two complicated fusion techniques which are fuzzy logic and sequential fuzzy logic and one simple fusion technique which is max voting. Lastly the rule-based decision is applied to specify whether the signature is genuine or not. Second, the improved signature verification system is extended with the high performance Hitachi system. This biometric based system can be realized in many real world and web based applications where there is a need for higher security and robust identification.