A Light weight and Hybrid Deep Learning Model based Online Signature Verification

Anoushka Doctor; Chandra Sekhar V.; Prerana Mukherjee; Viswanath Pulabaigiri

arxiv: 1907.04061 · v1 · pith:UALOJL3Znew · submitted 2019-07-09 · 💻 cs.CV · cs.LG

A Light weight and Hybrid Deep Learning Model based Online Signature Verification

Chandra Sekhar V. , Anoushka Doctor , Prerana Mukherjee , Viswanath Pulabaigiri This is my paper

Pith reviewed 2026-05-25 00:34 UTC · model grok-4.3

classification 💻 cs.CV cs.LG

keywords online signature verificationCNN-LSTM hybriddimensionality reductionMCYT datasetSUSIG datasetSVC datasetone-sample trainingdeep learning

0 comments

The pith

A hybrid CNN-LSTM model with dimensionality reduction verifies online signatures from one training sample.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a dimensionality reduction technique paired with a CNN-LSTM hybrid model for online signature verification. The goal is to create an efficient system that avoids the high training and storage costs of deeper networks while handling the common problem of scarce training examples. Experiments on the MCYT, SUSIG, and SVC datasets show the approach reaches higher accuracy than prior methods across forgery categories, even when trained on a single genuine signature. A sympathetic reader would care because practical verification systems often operate under tight data and compute limits yet still need reliable forgery detection.

Core claim

The authors establish that an efficient dimensionality reduction technique followed by a CNN-LSTM hybrid architecture delivers state-of-the-art performance in online signature verification on the MCYT, SUSIG, and SVC datasets, maintaining strong accuracy with as few as one training sample.

What carries the argument

Hybrid CNN-LSTM architecture applied after dimensionality reduction of signature features for genuine-versus-forged classification.

If this is right

The model outperforms prior methods on the MCYT, SUSIG, and SVC datasets in multiple forgery categories.
It maintains reliable performance when trained on only one genuine signature.
The reduced feature count lowers storage and computation demands for deployment on constrained devices.
State-of-the-art results hold across the three public datasets without requiring large training sets.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same reduction-plus-hybrid pattern could be tested on other biometric sequence tasks that face limited labeled examples.
The approach points toward lighter models for real-time verification on mobile hardware, though the paper does not report such tests.
Independent replication on a held-out dataset would clarify whether the gains depend on the specific preprocessing used in the three evaluated collections.

Load-bearing premise

The dimensionality reduction preserves enough information to separate genuine signatures from forgeries.

What would settle it

Training the model on one sample from the SVC dataset and obtaining verification accuracy well below the reported state-of-the-art levels on that dataset would falsify the central claim.

read the original abstract

The augmented usage of deep learning-based models for various AI related problems are as a result of modern architectures of deeper length and the availability of voluminous interpreted datasets. The models based on these architectures require huge training and storage cost, which makes them inefficient to use in critical applications like online signature verification (OSV) and to deploy in resource constraint devices. As a solution, in this work, our contribution is two-fold. 1) An efficient dimensionality reduction technique, to reduce the number of features to be considered and 2) a state-of-the-art model CNN-LSTM based hybrid architecture for online signature verification. Thorough experiments on the publicly available datasets MCYT, SUSIG, SVC confirms that the proposed model achieves better accuracy even with as low as one training sample. The proposed models yield state-of-the-art performance in various categories of all the three datasets.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper combines a CNN-LSTM hybrid with dimensionality reduction for online signature verification and claims solid accuracy from one training sample, but the abstract leaves the one-shot mechanism and experimental controls unaddressed.

read the letter

The core contribution here is a lightweight hybrid CNN-LSTM architecture paired with a dimensionality reduction step, aimed at online signature verification on resource-limited hardware. They test it on the usual public sets (MCYT, SUSIG, SVC) and report better accuracy than prior work even when limited to a single genuine signature per user, plus state-of-the-art numbers across categories. That combination is a reasonable incremental step for the OSV subfield, and the focus on low training cost and small model size is practically useful for deployment questions. The public datasets and the explicit efficiency goal are clear positives; anyone working on edge biometrics could pick up the architecture sketch without much trouble. The soft spot is the one-sample claim. Standard supervised training on a single example per class tends to overfit or memorize rather than generalize to skilled forgeries, and the abstract gives no indication of metric learning, pre-training, or other one-shot machinery that would make this feasible. The dimensionality reduction step is also presented without evidence that it keeps the features needed to separate genuine from forged signatures rather than just shrinking intra-user variance. Without methods details, baselines, or error bars, the SOTA assertion cannot be checked. This paper is mainly for applied researchers who need a compact model for signature checks on constrained devices. A reader already familiar with CNN-LSTM hybrids in biometrics will find the specific pairing and the one-sample numbers worth a quick look, but the work does not introduce new theory or a reusable framework. It deserves a serious referee if the full experiments section supplies the missing controls and comparisons; otherwise the central performance claim stays unconvincing. I would send it to review rather than desk-reject, mainly to see whether the one-sample results hold up under scrutiny.

Referee Report

3 major / 2 minor

Summary. The paper proposes a dimensionality reduction technique combined with a lightweight hybrid CNN-LSTM architecture for online signature verification. It reports that the model attains state-of-the-art accuracy on the public MCYT, SUSIG and SVC datasets even when trained with a single genuine signature per user.

Significance. A verified result would be significant for resource-constrained biometric applications, as it would demonstrate that a hybrid CNN-LSTM can operate reliably from extremely limited per-user data after dimensionality reduction. The use of three standard public datasets is a positive factor for potential reproducibility.

major comments (3)

[Abstract / Experiments] Abstract and Experiments section: the central claim that the model 'achieves better accuracy even with as low as one training sample' and yields SOTA performance is unsupported by any description of the training procedure, loss function, or one-shot mechanism; standard supervised training on a single sample would be expected to collapse to memorization unless transfer learning, metric learning or pre-training is employed, none of which is stated.
[Abstract] Abstract: no evaluation protocol, baseline methods, error bars, or cross-validation details are supplied for the reported accuracy figures on MCYT, SUSIG and SVC; without these the SOTA claim cannot be assessed and is load-bearing for the paper's contribution.
[Method / Dimensionality Reduction] Dimensionality reduction step: the manuscript must demonstrate that the reduced feature set retains forgery-discriminative information rather than merely lowering intra-user variance; no such analysis or ablation is referenced.

minor comments (2)

[Method] Notation for the hybrid CNN-LSTM layers and the exact form of the dimensionality reduction operator should be defined with equations.
[Experiments] Tables reporting accuracy should include the number of training samples used per user and the precise train/test split protocol.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight areas where additional clarity and analysis will strengthen the manuscript. We address each major comment below and will revise the paper to incorporate the requested details and experiments.

read point-by-point responses

Referee: [Abstract / Experiments] Abstract and Experiments section: the central claim that the model 'achieves better accuracy even with as low as one training sample' and yields SOTA performance is unsupported by any description of the training procedure, loss function, or one-shot mechanism; standard supervised training on a single sample would be expected to collapse to memorization unless transfer learning, metric learning or pre-training is employed, none of which is stated.

Authors: We agree that the manuscript does not sufficiently describe the training procedure, loss function, or one-shot mechanism supporting the single-sample claim. The current text is insufficient to substantiate how memorization is avoided. In the revised version we will add a dedicated subsection detailing the training protocol, the loss function employed, and the mechanism (including any pre-training or metric-learning components) that enables reliable performance from one sample. revision: yes
Referee: [Abstract] Abstract: no evaluation protocol, baseline methods, error bars, or cross-validation details are supplied for the reported accuracy figures on MCYT, SUSIG and SVC; without these the SOTA claim cannot be assessed and is load-bearing for the paper's contribution.

Authors: We acknowledge that the abstract and experiments section lack explicit statements of the evaluation protocol, baselines, error bars, and cross-validation procedure. These omissions prevent proper assessment of the reported figures. The revised manuscript will include a clear description of the protocol (including how the single-sample setting is constructed), the baseline methods, statistical error bars from repeated runs, and cross-validation details. revision: yes
Referee: [Method / Dimensionality Reduction] Dimensionality reduction step: the manuscript must demonstrate that the reduced feature set retains forgery-discriminative information rather than merely lowering intra-user variance; no such analysis or ablation is referenced.

Authors: We agree that an explicit demonstration is required to show the reduced features preserve forgery-discriminative information. The manuscript currently provides no ablation or analysis on this point. In the revision we will add an ablation study comparing verification performance before and after reduction, together with an analysis (e.g., feature variance or discriminative power metrics) confirming that forgery-related information is retained. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical model evaluation on public benchmarks

full rationale

The paper proposes a dimensionality-reduction step plus a CNN-LSTM hybrid and reports classification accuracies on the public MCYT, SUSIG and SVC datasets. No equations, uniqueness theorems, or fitted-parameter predictions are presented; performance numbers are obtained by standard train/test splits on external data. The central claim therefore rests on observable empirical outcomes rather than any self-referential reduction or self-citation chain, satisfying the default expectation of a non-circular empirical ML study.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract provides no mathematical formulation, so no free parameters, axioms, or invented entities can be identified.

pith-pipeline@v0.9.0 · 5687 in / 1151 out tokens · 31832 ms · 2026-05-25T00:34:59.394333+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

An efficient dimensionality reduction technique... K-Means clustering... Mean absolute difference (MAD)... CNN-LSTM based hybrid architecture
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

achieves better accuracy even with as low as one training sample... state-of-the-art performance

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

[1]

Abbas, Enhanced DTW based on- line signature verification

M.I.Khalil, M.Moustafa and H.M. Abbas, Enhanced DTW based on- line signature verification. In: Proceedings of the 16th IEEE InternationalConference on Image Processing (ICIP), 2009

work page 2009
[2]

Improving the enrollment in dynamic signature verfication with synthetic samples,

J. Galbally, J. Fiérrez, M. Diaz, and J. O.Garcia, “Improving the enrollment in dynamic signature verfication with synthetic samples,” in IntConf on DocAnal. Recognit. (ICDAR), p p. 1295 – 1299,Barcelona, Spain, 2009

work page 2009
[3]

Cancelable templates for sequence-based biometrics with application to on-line signature recognition,

E. Maiorana, P. Campisi, J. Fierrez, J. O.Garcia, and A. Neri, “Cancelable templates for sequence-based biometrics with application to on-line signature recognition,” IEEE Trans. Syst., Man, Cybern. A, Syst., Humans, vol. 40, no. 3, pp. 525–538, May 2010

work page 2010
[4]

Rashidi , A

S. Rashidi , A. Fallah , F. Towhidkhah , Authentication based on pole- zero models of signature velocity, Journal of Medical Signals Sens, vol 3, pp:195–208,2013

work page 2013
[5]

Barkoula, G

K. Barkoula, G. Economou and S. Fotopoulos, Online signature verification based on signatures turning angle representation using longest common subsequence matching, Int Journal of Doc Analysis and Recognition, vol 16, pp:261–272 , 2013

work page 2013
[6]

Online signature verification on mobile devices,

N. Sae-Bae and N. Memon, “Online signature verification on mobile devices,” IEEE Trans. Inf. Forensics Security, vol. 9, no. 6, pp. 933 – 947, Jun. 2014

work page 2014
[7]

Online signature verification using segment -level fuzzy modelling,

A. Q. Ansari, M. Hanmandlu, J. Kour, and A. Singh, “Online signature verification using segment -level fuzzy modelling,” IET Biometrics, vol. 3, no. 3, pp. 113–127, Sep. 2014

work page 2014
[8]

Online signature verification based on DCT and sparse representation,

Y. Liu, Z. Yang, and L. Yang, “Online signature verification based on DCT and sparse representation,” IEEE Trans. Cybern., vol. 45, no. 11, pp. 2498–2511, Nov. 2015

work page 2015
[9]

Pirlo, V

G. Pirlo, V. Cuccovillo, M. Diaz -Cabrera,D. Impedovo and P. Mignone, Multidomain Verification of Dynamic Signatures Using Local Stability Analysis, IEEE Transon Human -Machine Systems, vol 45, Dec. 2015

work page 2015
[10]

Towards an automatic on -line signature verifier using only one reference per signer,

M. Diaz, A. Fischer, R. Plamondon, and M. A. Ferrer, “Towards an automatic on -line signature verifier using only one reference per signer,” in Int. Conf. Document Anal. Recognit. (ICDAR), Tunis, Tunisia, pp. 631–635,2015

work page 2015
[11]

Pattern Recognition Letters,vol 84, pp:22–28, 2016

A.Sharma and S.Sundaram, An enhanced contextual dtw based system for online signature verification using vector quantization. Pattern Recognition Letters,vol 84, pp:22–28, 2016

work page 2016
[12]

Manjunatha, S

K.S. Manjunatha, S. Manjunath,D.S. Guru and M.T. Somashekara, 'Online signature verification based on writer dependent features and classifiers', Pattern Recognition Letters vol 80, pp: 129–136, 2016

work page 2016
[13]

S.Lai, L.Jin and W.Yang, Online Signature Verification Using Recurrent Neural Network and Length -Normalized Path Signature Descriptor, 14th IAPR IntConfon DocAna and Rec (ICDAR), 2017

work page 2017
[14]

2017, Kyoto, Japan

R.Tolosana, R.V.Rodriguez, J.Fierrez and J.Ortega-Garcia, Biometric Signature Verification Using Recurrent Neural Networks, 14th International Conference on Document Analysis and Recognition (ICDAR), Nov. 2017, Kyoto, Japan

work page 2017
[15]

M.Diaz, A.Fischer, M. A. Ferrer and R.Plamondon, Dynamic Signature Verification System Based on One Real Signature, IEEE Transactions On Cybernetics, vol 48, Jan 2018

work page 2018
[16]

L.Tang, W.Kang and Y.Fang, Information Divergence -Based Matching Strategy for Online Signature Verification, IEEE Transon Information Forensics and Security, vol 13, April 2018

work page 2018
[17]

A.Sharma and S.Sundaram, 'On the Exploration of Information From the DTW Cost Matrix for Online Signature Verification', IEEE Transactions on Cybernetics, vol 48, ffeb. 2018

work page 2018
[18]

67, Jan 2018

B.Kar, A.Mukherjee and P.K.Dutta, Stroke Point Warping -Based Reference Selection and Verification of Online Signature, IEEE Transactions On Instrumentation and Measurement, vol. 67, Jan 2018

work page 2018
[19]

L.Yang, Y.Cheng, X.Wang and Q.Liu, Online handwritten signature verification using feature weighting algorithm relief, Soft Computing, Vol 22, December 2018

work page 2018
[20]

S.Lai and L.Jin, Recurrent Adaptation Networks for Online Signature Verification, IEEE Trans on Information Forensics and Security, Nov 2018

work page 2018
[21]

R.Doroz, P.Kudlacik and P.Porwika, Online signature verification modeled by stability oriented reference signatures, vol 460 –461, pp: 151-171, Sep 2018

work page 2018
[22]

C.S.Vorugunti, D.S.Guru and P.Viswanath, An Efficient Onlin e Signature Verification Based on Feature Fusion and Interval Valued Representation of Writer Dependent Features, 'IEEE fifth International Conference on Identity, Security and Behavior Analysis (ISBA)', Jan 2019

work page 2019
[23]

R.Al-Hmouz, W.Pedrycz, K.Daqrouq, A.M orfeq and A.Al -Hmouz, Quantifying dynamic time warping distance using probabilistic model in verification of dynamic signatures, Elsevier -Soft Computing, vol 23, vol 23, pp 407–418, Jan 2019

work page 2019
[24]

L.He, H.Tan and Z.Huang, Online handwritten signature verification based on association of curvature and torsion feature with Hausdorff distance, springer : Multimedia Tools and Applications, Jan 2019. [39]. S.Yuan, X.Wu and Y.Xiang, Incorporating pre-training in long short- term memory networks for tweet class ification, Social Network Analysis and Minin...

work page 2019
[25]

P.Zhang, J.Xue, C.Lan, W.Zeng, Z.Gao and N.Zheng, SAdding Attentiveness to the Neurons in Recurrent Neural Networks, European Conference on Computer Vision, Sep 2018, Germany. [45]. R.Tolosana, R. Vera -Rodriguez, J.Fierrez, and J.Ortega -Garcia, Exploring Recurrent Neural Networks for On -Line Handwritten Signature Biometrics, IEEE Access, vol 6, pp:5128...

work page 2018

[1] [1]

Abbas, Enhanced DTW based on- line signature verification

M.I.Khalil, M.Moustafa and H.M. Abbas, Enhanced DTW based on- line signature verification. In: Proceedings of the 16th IEEE InternationalConference on Image Processing (ICIP), 2009

work page 2009

[2] [2]

Improving the enrollment in dynamic signature verfication with synthetic samples,

J. Galbally, J. Fiérrez, M. Diaz, and J. O.Garcia, “Improving the enrollment in dynamic signature verfication with synthetic samples,” in IntConf on DocAnal. Recognit. (ICDAR), p p. 1295 – 1299,Barcelona, Spain, 2009

work page 2009

[3] [3]

Cancelable templates for sequence-based biometrics with application to on-line signature recognition,

E. Maiorana, P. Campisi, J. Fierrez, J. O.Garcia, and A. Neri, “Cancelable templates for sequence-based biometrics with application to on-line signature recognition,” IEEE Trans. Syst., Man, Cybern. A, Syst., Humans, vol. 40, no. 3, pp. 525–538, May 2010

work page 2010

[4] [4]

Rashidi , A

S. Rashidi , A. Fallah , F. Towhidkhah , Authentication based on pole- zero models of signature velocity, Journal of Medical Signals Sens, vol 3, pp:195–208,2013

work page 2013

[5] [5]

Barkoula, G

K. Barkoula, G. Economou and S. Fotopoulos, Online signature verification based on signatures turning angle representation using longest common subsequence matching, Int Journal of Doc Analysis and Recognition, vol 16, pp:261–272 , 2013

work page 2013

[6] [6]

Online signature verification on mobile devices,

N. Sae-Bae and N. Memon, “Online signature verification on mobile devices,” IEEE Trans. Inf. Forensics Security, vol. 9, no. 6, pp. 933 – 947, Jun. 2014

work page 2014

[7] [7]

Online signature verification using segment -level fuzzy modelling,

A. Q. Ansari, M. Hanmandlu, J. Kour, and A. Singh, “Online signature verification using segment -level fuzzy modelling,” IET Biometrics, vol. 3, no. 3, pp. 113–127, Sep. 2014

work page 2014

[8] [8]

Online signature verification based on DCT and sparse representation,

Y. Liu, Z. Yang, and L. Yang, “Online signature verification based on DCT and sparse representation,” IEEE Trans. Cybern., vol. 45, no. 11, pp. 2498–2511, Nov. 2015

work page 2015

[9] [9]

Pirlo, V

G. Pirlo, V. Cuccovillo, M. Diaz -Cabrera,D. Impedovo and P. Mignone, Multidomain Verification of Dynamic Signatures Using Local Stability Analysis, IEEE Transon Human -Machine Systems, vol 45, Dec. 2015

work page 2015

[10] [10]

Towards an automatic on -line signature verifier using only one reference per signer,

M. Diaz, A. Fischer, R. Plamondon, and M. A. Ferrer, “Towards an automatic on -line signature verifier using only one reference per signer,” in Int. Conf. Document Anal. Recognit. (ICDAR), Tunis, Tunisia, pp. 631–635,2015

work page 2015

[11] [11]

Pattern Recognition Letters,vol 84, pp:22–28, 2016

A.Sharma and S.Sundaram, An enhanced contextual dtw based system for online signature verification using vector quantization. Pattern Recognition Letters,vol 84, pp:22–28, 2016

work page 2016

[12] [12]

Manjunatha, S

K.S. Manjunatha, S. Manjunath,D.S. Guru and M.T. Somashekara, 'Online signature verification based on writer dependent features and classifiers', Pattern Recognition Letters vol 80, pp: 129–136, 2016

work page 2016

[13] [13]

S.Lai, L.Jin and W.Yang, Online Signature Verification Using Recurrent Neural Network and Length -Normalized Path Signature Descriptor, 14th IAPR IntConfon DocAna and Rec (ICDAR), 2017

work page 2017

[14] [14]

2017, Kyoto, Japan

R.Tolosana, R.V.Rodriguez, J.Fierrez and J.Ortega-Garcia, Biometric Signature Verification Using Recurrent Neural Networks, 14th International Conference on Document Analysis and Recognition (ICDAR), Nov. 2017, Kyoto, Japan

work page 2017

[15] [15]

M.Diaz, A.Fischer, M. A. Ferrer and R.Plamondon, Dynamic Signature Verification System Based on One Real Signature, IEEE Transactions On Cybernetics, vol 48, Jan 2018

work page 2018

[16] [16]

L.Tang, W.Kang and Y.Fang, Information Divergence -Based Matching Strategy for Online Signature Verification, IEEE Transon Information Forensics and Security, vol 13, April 2018

work page 2018

[17] [17]

A.Sharma and S.Sundaram, 'On the Exploration of Information From the DTW Cost Matrix for Online Signature Verification', IEEE Transactions on Cybernetics, vol 48, ffeb. 2018

work page 2018

[18] [18]

67, Jan 2018

B.Kar, A.Mukherjee and P.K.Dutta, Stroke Point Warping -Based Reference Selection and Verification of Online Signature, IEEE Transactions On Instrumentation and Measurement, vol. 67, Jan 2018

work page 2018

[19] [19]

L.Yang, Y.Cheng, X.Wang and Q.Liu, Online handwritten signature verification using feature weighting algorithm relief, Soft Computing, Vol 22, December 2018

work page 2018

[20] [20]

S.Lai and L.Jin, Recurrent Adaptation Networks for Online Signature Verification, IEEE Trans on Information Forensics and Security, Nov 2018

work page 2018

[21] [21]

R.Doroz, P.Kudlacik and P.Porwika, Online signature verification modeled by stability oriented reference signatures, vol 460 –461, pp: 151-171, Sep 2018

work page 2018

[22] [22]

C.S.Vorugunti, D.S.Guru and P.Viswanath, An Efficient Onlin e Signature Verification Based on Feature Fusion and Interval Valued Representation of Writer Dependent Features, 'IEEE fifth International Conference on Identity, Security and Behavior Analysis (ISBA)', Jan 2019

work page 2019

[23] [23]

R.Al-Hmouz, W.Pedrycz, K.Daqrouq, A.M orfeq and A.Al -Hmouz, Quantifying dynamic time warping distance using probabilistic model in verification of dynamic signatures, Elsevier -Soft Computing, vol 23, vol 23, pp 407–418, Jan 2019

work page 2019

[24] [24]

L.He, H.Tan and Z.Huang, Online handwritten signature verification based on association of curvature and torsion feature with Hausdorff distance, springer : Multimedia Tools and Applications, Jan 2019. [39]. S.Yuan, X.Wu and Y.Xiang, Incorporating pre-training in long short- term memory networks for tweet class ification, Social Network Analysis and Minin...

work page 2019

[25] [25]

P.Zhang, J.Xue, C.Lan, W.Zeng, Z.Gao and N.Zheng, SAdding Attentiveness to the Neurons in Recurrent Neural Networks, European Conference on Computer Vision, Sep 2018, Germany. [45]. R.Tolosana, R. Vera -Rodriguez, J.Fierrez, and J.Ortega -Garcia, Exploring Recurrent Neural Networks for On -Line Handwritten Signature Biometrics, IEEE Access, vol 6, pp:5128...

work page 2018