Writer-Independent Signature Verification with Enhanced AlexNet and Preprocessing Analysis

Offline signature verification plays a crucial role in ensuring document authenticity, particularly in sectors reliant on handwritten signatures. This study introduces a Writer-Independent Signature Verification (WISV) system utilizing a Siamese neural network to compare signature pairs based on their cosine similarity. By preprocessing signature images into 400×400 pixel grayscale inputs, the model leverages convolutional neural networks (CNN) for feature extraction. The paper further evaluates the impact of data augmentation and aspect ratio preservation on model performance. Experimental results highlight that preserving input feature integrity during resizing enhances verification accuracy, providing a robust solution for signature verification.