Signature image identification using hybrid backpropagation with firefly algorithm and simulated annealing

Abstract

Signature pattern identification is a process of identifying pattern recognition because the signature is the primary mechanism for the authentication and authorization process in legal transactions. In this study, the identification of signature images using hybrid backpropagation with firefly algorithm and simulated annealing. There are three main stages in the backpropagation training method, namely feedforward, backpropagation of error, and updating weights and bias. Firefly algorithm and simulated annealing replace the backpropagation training process at the backpropagation of error stage and the weight and bias update stage, while for feedforward still use the existing algorithms in backpropagation training. The stages in the signature image identification process include image processing, namely the grayscale process, binary image, segmentation process, training process, and validation test process. Based on the results of the training process, the best weights and biases are obtained with a mean square error value of 0.00608. The results of the signature image identification show that the system has been able to recognize the image pattern well with a percentage of 93%.