COMPARATIVE ANALYSIS OF NEURAL NETWORK MODELS FOR HANDWRITTEN SIGNATURE DYNAMICS RECOGNITION

Abstract

In this paper, we investigated the application of neural networks for handwritten signature recognition taking into account the dynamics of their input. We considered three types of architectures: fully connected, LSTM, and convolutional networks. We analyzed the influence of key hyperparameters, such as the number of layers, activation function, optimizer, and batch size, on the classification accuracy. The experiments were performed on the MCYT Signature 100 dataset, which contains signatures of 100 authors and their forgeries. The results show that CNN and LSTM networks achieve the best accuracy with proper hyperparameter tuning and data preprocessing, including normalization and interpolation. The proposed method has practical value for automating signature verification and can be applied in the work of security systems and electronic document management.