A Writer-Independent Approach for Offline Signature Verification using Deep Convolutional Neural Networks Features

Abstract

The use of features extracted using a deep convolutional neural network (CNN) combined with a writer-dependent (WD) SVM classifier resulted in significant improvement in performance of handwritten signature verification (HSV) when compared to the previous state-of-the-art methods. In this work it is investigated whether the use of these CNN features provide good results in a writer-independent (WI) HSV context, based on the dichotomy transformation combined with the use of an SVM writer-independent classifier. The experiments performed in the Brazilian and GPDS datasets show that (i) the proposed approach outperformed other WI-HSV methods from the literature, (ii) in the global threshold scenario, the proposed approach was able to outperform the writer-dependent method with CNN features in the Brazilian dataset, (iii) in an user threshold scenario, the results are similar to those obtained by the writer-dependent method with CNN features.

A white-box analysis on the writer-independent dichotomy transformation applied to offline handwritten signature verification

“…In this paper, the SVM is used as writer-independent classifier with the following settings: RBF kernel, γ = 2 −11 and C = 1.0 (C and γ were selected based on a grid search: C grid = {0.0001, 0.001, 0.01, 0.1, 1, 10, 100} and γ grid = {2 −11 , 0.0001, 0.001, 0.01, 0.1, 1, 10, 100}). The signed distance of the samples to the classifier's hyperplane are used as classifiers output (Hafemann et al, 2017a;Souza et al, 2018b).…”

“…We have previously shown in (Souza et al, 2018b) In this section we present the results on the GPDS-300 exploitation set, comparing the results with the state-of-the-art. Table 7 contains both the comparison with the state of the art methods for the GPDS-300 dataset and also the results obtained by the WI-SVMs (with and without the CNN prototype selection).…”

“…The Dichotomy Transformation has already been used in various contexts, such as: bird species identification Zottesso et al (2018), forest species recognition Martins et al (2015), writer identification Bertolini et al (2016) and also for handwritten signature verification Rivard et al (2013); Eskander et al (2013); Souza et al (2018b).…”

“…The dichotomy transformation (DT) can be applied to deal with some of these challenges and therefore facilitate the signature verification task. The samples in the dissimilarity space (DS) generated by the dichotomy transformation are formed through the pairwise comparisons of signatures (a questioned and a reference signature) in the feature space (Souza et al, 2018b). Thus, having a good feature representation of the signatures is very important for DT to work (Bertolini et al, 2010).…”

Meta-Learning for Fast Classifier Adaptation to New Users of Signature Verification Systems

“…Table IV compares our results with the state-of-the-art. We observe an improved performance compared to other WI systems, achieving 5.16% EER (global τ ) with 5 reference signatures, compared to 9.05% from [30]. Comparing to WD systems, we observed similar performance in some scenarios (5 reference signatures), and worse results otherwise.…”