DWSCNN Online Signature Verification Algorithm Based on CAE-MV Feature Dimensionality Reduction
Jianbin Zheng,Ziyao Chen,Liping Huang,Yifan Gao,Xiangxiang Yu,Hui Wang,Jiamei Yang,Yu Wang
DOI: https://doi.org/10.1109/access.2024.3355449
IF: 3.9
2024-02-16
IEEE Access
Abstract:In recent years, with the rapid advancements in deep learning technologies, particularly deep neural networks, signature verification has seen significant improvements in accuracy. Despite the significant progress made in using deep learning technologies, there remains several challenges that affect their practical application, such as insufficient feature extraction, long neural network computation time, and high resource consumption. In this paper, an online handwritten signature verification algorithm is proposed, which mainly uses the CAE-MV-based feature dimensionality reduction method to compress and select features of the original signature data to construct a signature feature set. The Depth-wise Separable Convolutional Neural Network(DWSCNN) based on Depth-wise separable convolution is used to classify and verify the signature feature set. Compared with CNN, the DWSCNN can significantly reduce the number of neural network parameters while the classification effect is not much different, thus greatly shortening the running time and resource occupation. Through the analysis of the verification results on two publicly available signature databases, MCYT-100 and SVC-2004, the proposed algorithm improves the accuracy, reduces the False Acceptance Rate (FAR) and False Rejection Rate (FRR). The average verification accuracy reaches 98.92%, which is superior to the current mainstream online signature verification frameworks, demonstrating the effectiveness and efficiency of the proposed framework.
computer science, information systems,telecommunications,engineering, electrical & electronic