Online signature verification based on feature representation

Abstract

Abstract-Biometrics systems have been used in a wide range of applications and have improved people authentication. Signature verification is one of the most common biometric methods with techniques that employ various specifications of a signature. Recently, deep learning has achieved great success in many fields, such as image, sounds and text processing. In this paper, deep learning method has been used for feature extraction and feature selection, which has enormous impact on the accuracy of signature verification. This paper presents a method based on self-taught learning, in which a sparse autoencoder attempts to learn discriminative features of signatures from a large unlabeled signature dataset. Then, the features learned are employed to present users' signatures by creating a model for each user based on user genuine signatures. Finally, users' signatures are classified using a one-class classifier. The proposed method is independent on signature datasets thanks to self-taught learning. The features have been learned from 17,500 signatures (ATVS dataset) and verification process of the proposed system is evaluated on SVC2004 and SUSIG signature datasets, which contain genuine and skilled forgery signatures. The experimental results indicate significant error reduction and accuracy enhancement in comparison with state of the art counterparts.

A novel approach for Finger Vein verification based on self-taught learning

“…Although in local threshold, system must choose one threshold per user. However respect to generality and complexity the global threshold has a better performance rather than local threshold, but in comparison of accuracy term it is worse [12].…”

Forged Signature Distinction Using Convolutional Neural Network for Feature Extraction

“…To improve the accuracy of signature verification, some studies [8][9][10] utilize machine learning techniques, which are one of the most noteworthy technologies. Similar to the process in feature-based signature verification, they use descriptive features of signatures for building a subject model.…”

“…Iranmanesh et al [9] exploited MLP for signature verification, and demonstrated an average accuracy of 82.42% for recognizing the subject; however, this method could not distinguish between imitated and original signatures. As mentioned previously, M. Fayyza et al [10] used AE, the one-class model, to distinguish imitated signatures. They derived the (x, y) coordinates, sign changes (dx/dt), (dy/dt), pen pressure, and pen angle as features.…”

“…Early studies [4][5][6][7][8][9][10] primarily focused on verifying signatures drawn by a stylus. Finger-drawn signature verification [3,[11][12][13] is relatively not well researched yet; however, it is currently actively discussed because most smartphones do not have a stylus, and finger signing is more convenient.…”

3DAirSig: A Framework for Enabling In-Air Signatures Using a Multi-Modal Depth Sensor

“…The artificial neural network (ANN), support vector machine (SVM), and pixel matching technique (PMT) are famous classification algorithms, which have been used by offline methods. On the other hand, in online methods, e-signatures are taken on a touch device (e.g., tablet or pad) using an e-pen or finger movement on a digital screen [ 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 ]. These methods are difficult to forge due to various dynamic features, such as velocity, acceleration, and pen pressure.…”