Detection of Offline Handwritten Signature Forgery Using InceptionV4 with Sobel Edge Detection

Handwritten signature forgery detection is essential for authenticating signatures to address the risks of document falsification, identity theft, and financial fraud from signature forgeries. This study presents a reliable approach to accurately classify genuine and forged signatures using deep learning and image pre-processing techniques. Specifically, the proposed method utilizes InceptionV4 with Sobel edge detection. InceptionV4, a deep convolutional neural network, is used to extract features from signature images, while Sobel edge detection enhances these features. By applying Sobel edge detection during the image pre-processing stage, we emphasize the edges of the signatures, which aids in the classification process. Furthermore, we implemented data augmentation techniques to effectively generate additional datasets, overcoming the significant limitations of the lack of custom signature datasets. By integrating Inception V4’s capabilities and Sobel edge detection, the model achieved a high accuracy of 96.19% and an F1-score of 96.18%. These results demonstrate the potential and effectiveness of the proposed approach in detecting signature forgeries.