A Hybrid Features for Signature Recognition Using Neural Network


In automatic personal recognition systems, biometric features is used as recognition measure based on biological traits such as face, iris, fingerprint, etc…or gait, signature which is considered behavioral characteristics. Signature verification is one of the authentication methods which can provide security at maintenance and low cost. The most essential and challenging stage of any off-line signature system is feature extraction stage. The accuracy and robust of the recognition system depends basically on the usefulness of the signature features extracted by this system. If the extracted features from a signature's image doesn't robust this will cause to higher verification error-rates especially for skilled forgeries in hacker the system. In this paper, we present a new offline handwritten signature recognition system based on combination of global with Statistical and GLCM (Grey Level Co-occurrence Matrix) features using neural network as classifier tool. The global, Statistical and GLCM features are combined to consist a vector of 14 features for the authentication of the signature. Verification of signatures is decided using neural network. The experimental results obtained by using a database of 7 individuals’ signatures. A total number of 70 images are collected with 10 signatures for each person, 5 of the signatures are used in training phase, and the remaining 5 signatures are used in testing phase. In this proposed method the results show 100% recognition accuracy for training and 97.1% recognition accuracy for the testing.