Analyzing features learned for Offline Signature Verification using Deep CNNs

Abstract

Abstract-Research on Offline Handwritten Signature Verification explored a large variety of handcrafted feature extractors, ranging from graphology, texture descriptors to interest points. In spite of advancements in the last decades, performance of such systems is still far from optimal when we test the systems against skilled forgeries -signature forgeries that target a particular individual. In previous research, we proposed a formulation of the problem to learn features from data (signature images) in a Writer-Independent format, using Deep Convolutional Neural Networks (CNNs), seeking to improve performance on the task. In this research, we push further the performance of such method, exploring a range of architectures, and obtaining a large improvement in state-of-the-art performance on the GPDS dataset, the largest publicly available dataset on the task. In the GPDS-160 dataset, we obtained an Equal Error Rate of 2.74%, compared to 6.97% in the best result published in literature (that used a combination of multiple classifiers). We also present a visual analysis of the feature space learned by the model, and an analysis of the errors made by the classifier. Our analysis shows that the model is very effective in separating signatures that have a different global appearance, while being particularly vulnerable to forgeries that very closely resemble genuine signatures, even if their line quality is bad, which is the case of slowly-traced forgeries.

A Comprehensive Study of Sparse Representation Techniques for Offline Signature Verification

“…Specifically, the authors in [27] introduced a formulation for learning features from genuine signatures by a development dataset, and used them in order to train writer dependent classifiers to another set of users. In [28] the authors obtained state-of-the-art results on several GPDS datasets using CNN architecture and in [29] they demonstrated a novel formulation that leverages knowledge of skilled forgeries for feature learning. In addition, the authors in [8] responded to the fixed size input constraint of the neural network by learning a fixed-sized representation from variable sized signature images with the integration of a spatial pyramid pooling layer.…”

Fixed-sized representation learning from offline handwritten signatures of different sizes

“…In previous work [9] we performed data augmentation by performing random crops of the input images. We adopt the same protocol for the "Fixed size" training.…”

“…In all cases, we trained a binary SVM, with an RBF kernel. We used the same hyperparameters as previous research [9]: C = 1 and γ = 2 −11 , that were selected using a subset of the GPDS validation set. In this paper we did not explore optimizing these hyperparameters for each dataset (or even each user), but rather keep the same set of parameters for comparison with previous work.…”

Learning features for offline handwritten signature verification using deep convolutional neural networks

“…In [10], we introduced the formulation to learn features from genuine signatures from a development dataset, using them to train Writer-Dependent classifiers to another set of users. In [11], we analyzed the learned feature space and optimized the CNN architecture, obtaining state-of-the-art results on GPDS. The present work includes this formulation of the problem for completeness, with additional experiments on two other datasets (MCYT and CEDAR), a clearer explanation of the method and the experimental protocol, as well as the novel formulation that leverages knowledge of skilled forgeries for feature learning.…”