Table 1 Detailed analysis of existing related work.
From: Handwriting identification and verification using artificial intelligence-assisted textural features
Ref. paper | Main focus | Findings | Limitations |
|---|---|---|---|
AVN for authenticating signatures | 94% accuracy in verification and feature extraction | Slower performance with less variation | |
Transformer model for analysis of signatures | 95.4% accuracy and 97.8% efficiency | Inadequate data about ET victims | |
CNN for identifying and detecting calligraphy | Verification accuracy of 96.8% | Many handwriting samples are necessary | |
Predicting handwriting position without supervision | 93.3% posture prediction accuracy | Challenges of dimensionality reduction | |
ShuffleNet CNN in order to identify characters | 99.50% accuracy in character recognition | Limited Sample size | |
Siamese neural network system for verifying signatures | 99.06% accuracy in verification, less time and effort | Needs large labelled data | |
Gender identification using ATP-DenseNet | Gender identification accuracy of 66.3% | Cropping difficulties | |
Data enhancement for OHR systems | Geometric technique, 97.2% recognition accuracy | Limited samples in some labels | |
A writer retrieval system based on SVM | improved performance and feasibility by 96.76% | Signature documents must be sorted | |
Online signature verification with barcodes | 97.9%verification accuracy with wavelet selection | Wavelet selection was optimal | |
Signature augmentation parameter optimisation | Parameter optimization enhanced performance by 95.6% | Compact clusters were not produced | |
BEM signature steganography | Verification accuracy of 84.38%, reduced computation time | Instability of lossy compression | |
For online systems, the SVSV technique was employed | Signature verification accuracy of 98.4%, sound and vibration | Robustness inconsistence in a variety of positions | |
Forensic students can benefit from collaborative learning | Learning efficiency of 93.4%, mistakes and complexity reduced | Some qualities are problematic | |
Fusion of feature extraction for writer-independent | CCA analysis, LBP features, and 86% verification accuracy | Visual depiction is limited | |
Handwritten image identification using dual-fuzzy CNN | Model and approach that is both feasible and useful | Limited samples | |
Dynamic Signature Fuzzy Vault Scheme | Improved evaluation performance and efficiency | Inefficient use of time | |
two-level Haar wavelet transformation | Improved recognition accuracy, FAR (0.030%) and FRR (0.025%) | Occurrence of overlap |
