| Abstrakt: |
A handwritten signature is the most widely accepted method to authenticate an individual in banking, financial, business transactions, cheque processing, access control, and e-business due to its simplicity and distinctiveness. Many automated systems have been created to predict the authenticity of a signature. However, barely a few review papers provide a summary of the existing research on deep learning-based signature verification systems. An attempt is made to bring out the strengths and find gaps in handwritten signature verification in various dimensions considering both online and offline signatures. The primary objective of this comprehensive study is to present the most recent deep learning-based models for signature verification systems by putting emphasis on five different aspects: datasets, preprocessing techniques, feature extraction methods, machine learning-based verification models, and performance evaluation metrics. This review includes publications on both offline and online signature verification systems published between 2017 and 2022. This systematic review has discovered that recently, the deep learning-based neural network accomplished the most encouraging results for signature verification systems on public datasets. This comprehensive review revealed that recently, the deep learning-based neural network attained the most promising results for signature verification systems on public datasets. This article distinguishes itself from other reviews by revealing the twelve most significant research areas for researchers. (1) Intra-personal and inter-personal variability. (2) Few-shot Learning. (3) Offline signature recovery from online signature and vice versa. (4) Device Interoperability issues in the multi-device scenario. (5) Extracting signatures from Real World Document images. (6) Signature segmentation using Hyper Spectral Imaging. (7) Signatures as a means of retrieval of documents. (8) Multiscript Signature Verification Systems. (9) Automatic health state assessment. (10) Ensemble of Classifiers. (11) Synthetic Signature Generation. (12) Transfer Learning. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink