Signature forgery detection plays a crucial role in ensuring the authenticity of handwritten signatures, which are widely used in financial transactions, legal documentation, and personal identification. Traditional machine learning approaches have shown moderate success in this domain but are limited by their dependence on manual feature extraction and inability to capture subtle variations in signatures. In this paper, we propose a Siamese-based Convolutional Neural Network (CNN) with residual attention mechanisms to address these limitations. The Siamese network is designed to directly compare pairs of signature images, focusing on subtle differences between genuine and forged signatures, while the residual attention mechanism enhances the model’s ability to dynamically attend to key signature features. We evaluated our model on the ICDAR 2011 Signature Dataset, which contains both genuine and forged signatures from Dutch users. The proposed approach achieved an accuracy of 98.76%, outperforming traditional classifiers and standard CNN architectures. This work demonstrates the effectiveness of attention-based deep learning models in signature forgery detection and opens avenues for further research into more robust and generalized signature verification systems.