Enhancing Offline Handwritten Signature Identification with Pre-Trained CNN Architectures
Nader Ebrahimpour · 2019
Handwritten Signature Recognition (HSR) is a vital task in document authentication and verification systems. This paper proposes a novel approach for offline HSR leveraging pre-trained Convolutional Neural Network (CNN) models. CNNs have demonstrated remarkable performance in various computer vision tasks, including image recognition, making them suitable for HSR tasks. Our proposed method uses pre-trained CNN models trained on large-scale image datasets, such as ImageNet, to extract high-level features from handwritten signature images. By fine-tuning these pre-trained models on a dataset of offline handwritten signatures, we aim to transfer the learned knowledge to the task of HSR. We explore different pre-trained CNN architectures, such as MobileNet, ShuffleNet, ResNet, and EfficientNet, and investigate their performance in HSR tasks. Furthermore, we propose a signature verification system that combines the features extracted from pre-trained CNN models with Euclidean Distance (ED) metric to authenticate handwritten signatures. Experimental results on benchmark datasets demonstrate the effectiveness of our proposed approach in achieving state-of-the-art performance in offline HSR tasks.