|
| Title | Authors | Year | Modality | Characteristics | Performance metrics |
|
| From text to signatures: knowledge transfer for efficient deep feature learning in offline signature verification [31] | Dimitrios Tsourounis, Ilias Theodorakopoulos, Elias N. Zois, George Economou | 2022 | Offline | It uses handwritten text for the initial training of a convolutional neural network instead of handwritten signatures. This scheme uses three publicly available databases for signature verification namely CEDAR, MCYT-75, and GPDS300 gray | The average EER was found to be 2.5 approx |
|
| Offline signature verification using a region based deep metric learning network [26] | Li Liu, Linlin Huang, Fei Yin, Youbin Chen | 2021 | Offline | A region-based Deep Convolutional Siamese network using the metric learning method, which applies to both writer-dependent (WD) and writer-independent (WI) scenarios was proposed
Mutual Signature DenseNet (MSDN) is designed to extract features and learn the similarity measure from local regions | In WD EER = 6.74 and EER = 1.67 in WI using CEDAR database |
|
| CBCapsNet: a novel writer-independent offline signature verification model using a CNN-based architecture and capsule neural networks [27] | Ebrahim Parcham, Mahdi Ilbeygi, Mohammad Amini | 2021 | Offline | A combination of CNN and CapsNet is proposed to extract spatial features. The training method proposed uses two images simultaneously to train a single network to reduce parameters by 50%
This technique has been tested under CEDAR, GPDS300, GPDS Synthetic, Bengali, and Hindi datasets | FAR ranges from 0–9.45 and FRR ranges from 0–8.81 for CBCapsNet |
|
| A recurrent neural network based deep learning model for offline signature verification and recognition system [29] | Rajib Ghosh | 2021 | Offline | Uses RNN on a deep learning model. The significance of the proposed technique is it can be used for multiscript signatures. Various datasets like MCYT, GPDS300, CEDAR, BHSig260 Hindi, BHSig260 Bengali, GPDS synthetic etcetera were used to test the performance of the system | The average accuracy for different datasets and feature sets is around 94% |
|
| Learning the micro deformations by max-pooling for offline signature verification [30] | Yuchen Zheng, Brian Kenji Iwana, Muhammad Imran Malik, Sheraz Ahmed, Wataru Ohyama, Seiichi Uchida | 2021 | Offline | A signature verifier system based on micro deformations is proposed. The verifier is based on a convolutional neural network. The system has been tested against SyntheticGPDS, CEDAR, UTSig, and BHSig260 datasets | EER for random forgery ranges from 0 to 1.78 and for skilled forgery from 8.31 to 33.82 |
|
| Off-line handwritten signature verification using compositional synthetic generation of signatures and Siamese neural networks [17] | Victoria Ruiz, Ismael Linares, Angel Sanchez, Jose F. Velez | 2020 | Offline | A Siamese neural network is used and trained using samples from the GAVAB dataset and different synthetic data.
A combination of real and synthetic data gave a better performance. The system was tested using GPSSynthetic, MCYT, SigComp11, and CEDAR datasets | EER for different datasets ranges from 2.06 to 4.84 |
|
| Generation of duplicated off-line signature images for verification systems [4] | Diaz, M., Ferrer, M.A., Eskander, G.S., Sabourin, R. | 2017 | Offline | Based on cognitive inspired theory. A group of nonlinear and linear transformations is used to simulate the human spatial cognitive map and motor system to produce variability among the specimens. The datasets used are GPDS-300 and MCYT-75 | The performance of the system is highly dependent on the tunable parameters of the synthetic sample generator |
|
| Quantifying dynamic time warping distance using probabilistic model in verification of dynamic signatures [28] | Rami Al-Hmouz, Witold Pedrycz, Khaled Daqrouq, Ali Morfeq, Ahmed Al-Hmouz | 2017 | Online | Probabilistic time warping is proposed in this paper to reduce the EER. Two publicly available synthetic datasets are used for training purposes viz. SCV2004 and MYCT | EER on SCV2004 and MYCT100 are found to be 0.018 and 0.019 |
|
| Spline interpolation and deep neural networks as feature extractors for signature verification purposes [32] | Wei Wei, Qiao Ke, Dawid Połap, Marcin Wo´zniak | 2015 | Offline | Spline interpolation in addition to two neural networks has been used. Local features are extracted using interpolation methods and global features are verified using convolutional neural networks | Accuracy 87% |
|
