Deep Boltzmann machines for robust fingerprint spoofing attack detection

Abstract

Biometric systems have been widely considered as a synonym of security. However, in recent years, malicious people are violating them by presenting forged traits, such as gelatin fingers, to fool their capture sensors (spoofing attacks). To detect such frauds, methods based on traditional image descriptors have been developed, aiming liveness detection from the input data. However, due to their handcrafted approaches, most of them present low accuracy rates in challenging scenarios. In this work, we propose a novel method for fingerprint spoofing detection using the Deep Boltzmann Machines (DBM) for extraction of high-level features from the images. Such deep features are very discriminative, thus making complicated the task of forgery by attackers. Experiments show that the proposed method outperforms other state-of-the-art techniques, presenting high accuracy regarding attack detection.

Deep Features Extraction for Robust Fingerprint Spoofing Attack Detection

“…In this Section, we briefly describe an approach that we previously proposed in [5] for fingerprint spoofing detection, also based on a DBM for deep features extraction. Actually, in such previous work and different from the actual one, after learning the parameters of the DBM, a final layer was added at the top of such a structure with two softmax units, forming an MLP (Multilayer Perceptron) network, i.e., a complete classifier, to identify normal (class "0") or attack fingerprint patterns (class "1").…”

“…Basically, given an initial training set of grayscale fingerprint images, the first step consisted in the extraction of their relevant regions [5]. After that, resizing and database augmentation techniques were also applied to improve the network performance and avoid lack of data in training.…”

“…augmentation process: given each initial high-dimensional training fingerprint image, its ROI was detected, cropped, resized, flipped and 10 different patches were obtained based on the original and flipped ROI, by translating the position of the red rectangle in their four corners and central regions. This process was repeated for all training images and at the end, the patches generated from all of them served as an input to train the DBM [5].…”

“…As mentioned in [5], after preprocessing the images from the dataset and before performing the training of the DBM itself, given each grayscale fingerprint patch, it was needed to previously train a Gaussian-Bernoulli RBM (GB-RBM), that served as the interface between such real-valued patches and the Bernoulli-Bernoulli RBMs (BB-RBMs) that constituted the DBM. This previous step with the GB-RBM was called GB Preprocessing and is shown in Figure 5.…”

“…In this work, we propose a novel approach for spoofing detection in fingerprint recognition systems, an adaptation of the method we presented in [5], using deep features extracted from images by a probabilistic deep learning architecture: the Deep Boltzmann Machine (DBM) [6]. DBMs can deal with complex patterns efficiently and accurately since they extract and work with high-level features from the original data, being suitable for tasks in which the patterns should not be easily detected or forged.…”