Edson J. R. Justino

Off-line signature verification using HMM for random, simple and skilled forgeries

The problem of signature verification is in theory a pattern recognition task used to discriminate two classes, original and forgery signatures. Even after many efforts in order to develop new verification techniques for static signature verification, the influence of the forgery types has not been extensively studied. This paper reports the contribution to signature verification considering different forgery types in an HMM framework. The experiments have shown that the error rates of the simple and random forgery signatures are very closed. This reflects the real applications in which the simple forgeries represent the principal fraudulent case. In addition, the experiments show promising results in skilled forgery verification by using simple static and pseudodynamic features.

A comparison of SVM and HMM classifiers in the off-line signature verification

The SVM is a new classification technique in the field of statistical learning theory which has been applied with success in pattern recognition applications like face and speaker recognition, while the HMM has been found to be a powerful statistical technique which is applied to handwriting recognition and signature verification. This paper reports on a comparison of the two classifiers in off-line signature verification. For this purpose, an appropriate learning and testing protocol was created to observe the capability of the classifiers to absorb intrapersonal variability and highlight interpersonal similarity using random, simple and simulated forgeries.

Reducing forgeries in writer-independent off-line signature verification through ensemble of classifiers

In this work we address two important issues of off-line signature verification. The first one regards feature extraction. We introduce a new graphometric feature set that considers the curvature of the most important segments, perceptually speaking, of the signature. The idea is to simulate the shape of the signature by using Bezier curves and then extract features from these curves. The second important aspect is the use of an ensemble of classifiers based on graphometric features to improve the reliability of the classification, hence reducing the false acceptance. The ensemble was built using a standard genetic algorithm and different fitness functions were assessed to drive the search. Two different scenarios were considered in our experiments. In the former, we assume that only genuine signatures and random forgeries are available to guide the search. In the latter, on the other hand, we assume that simple and simulated forgeries also are available during the optimization of the ensemble. The pool of base classifiers is trained using only genuine signatures and random forgeries. Thorough experiments were conduct on a database composed of 100 writers and the results compare favorably.

Texture-based descriptors for writer identification and verification

Highlights? A segmentation free process for writer identification/verification. ? Evaluation of two texture descriptors (LBP and LPQ) for writer identification/verification. ? Evaluation of the dissimilarity-based approach for writer identification. ? Discussion about the number and size of the references for the dissimilarity-based approach. In this work, we discuss the use of texture descriptors to perform writer verification and identification. We use a classification scheme based on dissimilarity representation, which has been successfully applied to verification problems. Besides assessing two texture descriptors (local binary patterns and local phase quantization), we also address important issues related to the dissimilarity representation, such as the impact of the number of references used for verification and identification, how the framework performs on the problem of writer identification, and how the dissimilarity-based approach compares to other feature-based strategies. In order to meet these objectives, we carry out experiments on two different datasets, the Brazilian forensic letters database and the IAM database. Through a series of comprehensive experiments, we show that both LBP- and LPQ-based classifiers are able to surpass previous results reported in the literature for the verification problem by about 5 percentage points. For the identification problem, the proposed approach using LPQ features is able to achieve accuracies of 96.7% and 99.2% on the BFL and IAM and databases respectively.

An off-line signature verification method based on the questioned document expert's approach and a neural network classifier

In an off-line signature verification method based on personal models, an important issue is the number of genuine samples required to train the writers model. In a real application, we are usually quite limited in the number of samples we can use for training [Cha, S., 2001, Baltzakis, H. et al., 2001, Yingyong, Q. et al., 1994]. Classifiers like the neural network [Baltzakis, H. et al., 2001], the hidden Markov model [Justino, E.J.R. et al., 2001] and the support vector machine [Justino, E.J.R. et al., 2003] need a substantial number of samples to produce a robust model in the training phase. This paper reports on a global method based on only two classes of models, the genuine signature and the forgery. The main objective of this method is to reduce the number of signature samples required by each writer in the training phase. For this purpose, a set of graphometric features and a neural network (NN) classifier are used.

The interpersonal and intrapersonal variability influences on off-line signature verification using HMM

Off-line signature verification rests on the hypothesis that each writer has similarity among signature samples, with small distortion and scale variability. This kind of distortion represents intrapersonal variability. This paper reports interpersonal and intrapersonal variability influences in a software approach based on a hidden Markov model (HMM) classifier. The experiments have shown error rate variability considering different forgery types, random, simple and skilled. The mathematical approach and resulting software also report considerations in a real application problem.

Off-line signature verification using HMMs and cross-validation

We propose an HMM-based approach for off-line signature verification. One of the novelty aspects of our method lies in the ability to dynamically and automatically derive the various author-dependent parameters, required to set an optimal decision rule for the verification process. In this context, the cross-validation principle is used to derive not only the best HMM models, but also an optimal acceptation/rejection decision threshold for each author. This leads to a high discrimination between actual authors and impostors in the context of random forgeries. To quantitatively evaluate the generalization capabilities of our approach, we considered two conceptually different experimental tests carried out on two sets of 40 and 60 authors respectively, each author providing 40 signatures. The results obtained on these two sets show the robustness of our approach.

Writer verification using texture-based features

In this work, we propose a writer verification system that takes into account texture-based features and dissimilarity representation. Textures of the handwritings are created based on the inherent properties of the writer. Independent of the writing style, the proposed method reduces the spaces between lines, words, and characters, producing a texture that keeps the main features thus avoiding the complexity of segmentation. We also address an important issue of verification system, i.e., the number of writers used for training. Our experiments show that the number of writers do not have an important impact on the overall error rate, but it has an important role in reducing the false acceptance of the verification system. We show that the false acceptance decreases as the number of writers increases. Finally, the ROC curves produced by different classifiers trained with different texture descriptors are combined using the maximum likelihood analysis, producing a ROC combined classifier. A set of experiments on a database composed of 315 writers show the efficiency of the texture-based features and the ROC combination scheme. Experimental results report an overall error rate of about 4%. This performance compares to the state of the art. Besides, the combination scheme is able to considerably reduce the false-positive rates while maintaining the same true-positive rates.

An off-line signature verification system using hidden Markov model and cross-validation

The main objective is to present an off-line signature verification system. It is basically divided into three parts. The first demonstrates a pre-processing process, a segmentation process and a feature extraction process, in which the main aim is to obtain the maximum performance quality of the process of verification of random falsifications, in the false acceptance and false rejection concept. The second presents a learning process based on HMM, where the aim is obtaining the best model. That is, one that is capable of representing each writers signature, absorbing yet at the same time discriminating, at most the intrapersonal and interpersonal variation. The third presents a signature verification process that uses the models generated by the learning process without using any prior knowledge of test data, in other words, using an automatic derivation process of the decision thresholds.

Off-line Signature Verification Using Writer-Independent Approach

In this work we present a strategy for off-line signature verification. It takes into account a writer-independent model which reduces the pattern recognition problem to a 2-class problem, hence, makes it possible to build robust signature verification systems even when few signatures per writer are available. Receiver operating characteristic (ROC) curves are used to improve the performance of the proposed system . The contribution of this paper is two-fold. First of all, we analyze the impacts of choosing different fusion strategies to combine the partial decisions yielded by the SVM classifiers. Then ROC produced by different classifiers are combined using maximum likelihood analysis, producing an ROC combined classifier. Through comprehensive experiments on a database composed of 100 writers, we demonstrate that the ROC combined classifier based on the writer-independent approach can reduce considerably false rejection rate while keeping false acceptance rates at acceptable levels.