Inverse Biometrics: Generating Vascular Images from Binary Templates -- Downloads and Info

This work as a submussion to the IEEE Transactions on Biometrics is an extension of our previous work [Kauba20a], submitted to the IAPR/IEEE International Joint Conference on Biometrics (IJCB2020) with the title "Inverse Biometrics: Reconstructing Grayscale Finger Vein Images from Binary Features". In our current work we extended our approach to hand vein images as well. Hence, vascular pattern images (finger- as well as hand-vein images) are reconstructed from their feature-rich binary templates using our proposed deep-learning based alogrithm that has been investigated in the scope our previous work. As the paper is constrained in the number of page, all the necessary extra information, including setting-files and recognition result files can be downloaded from this site to comply with the principles of reproducible research. Furthermore, several in-detail DET plots and other results evaluations are available as well.

Abstract

In this work, we investigate the possibility of generating grayscale images of finger and hand vein patterns from their corresponding binary templates. This exercise would allow us to determine the invertibility of vascular templates, which has implications in biometric security and privacy. The transformation from binary features to a gray-scale image is accomplished using a Pix2Pix Convolutional Neural Network (CNN). The reversibility of 6 different types of binary features is evaluated using this CNN. Further, a number of experiments are conducted using 7 distinct finger-vein datasets and 3 hand-vein datasets. Results indicate that (a) it is possible to reconstruct the considered vascular images from their binary templates; (b) the reconstructed images can be used for biometric recognition purposes; (c) the CNN trained on one dataset can be successfully used for reconstructing images in a different dataset (cross-dataset reconstruction); and (d) the images reconstructed from one set of features can be successfully used to extract a different set of features for biometric recognition (cross-feature-set generalization). The results of this research further underscore the need for properly securing biometric templates, even if they are of binary nature.

References

[Kauba20b  ] Inverse Biometrics: Generating Vascular Images from Binary Templates Christof Kauba, Simon Kirchgasser, Vahid Mirjalili, Arun Ross, Andreas Uhl In T-BIOM, pp. 1-11, TODO, to appear 2021

[Kauba20a  ] Inverse Biometrics: Reconstructing Grayscale Finger Vein Images from Binary Features Christof Kauba, Simon Kirchgasser, Vahid Mirjalili, Arun Ross, Andreas Uhl In Proceedings of the IAPR/IEEE International Joint Conference on Biometrics (IJCB2020), pp. 1-8, Houston, Texas, USA, September 28 - October 1, accepted

Data Sets

PLUSVein-FV3 Finger Vein Data Set

The PLUSVein-FV3 Finger Vein Data Set (PLUSVein-FV3) is a publically available fingr vein data set. It contains palmar and dorsal images of 360 fingers from 60 different subjects (ring, middle and index finger from both hands) captured in one session with five samples per finger using two different variants of the same sensor: One utilizing NIR laser modules for illumination, the other one using NIR LEDs. Further information regarding the data set can be found by the following link:

PLUSVein-FV3 Data Set

University of Twente Finger Vascular Pattern Database

The University of Twente Finger Vascular Pattern (UTFVP) Database is a publically available finger vein data set. It contains six fingers (ring, middle and index finger from both hands) from 60 volunteers acquired in two sessions. Further information regarding the data set can be found by the following link:

UTFVP Download Page (external link)

Finger Vein USM (FV-USM) Database

The Finger Vein USM (FV-USM) Database is a publically available finger vein data set. The images in the database were collected from 123 volunteers comprising of 83 males and 40 females, who were staff and students of Universiti Sains Malaysia. Further information regarding the data set can be found by the following link:

FV-USM Download Page (external link)

The Chonbuk University Finger Vein Database

The Chonbuk University in South Korea used their prototype scanner (Chonbuk Proto) to establish the MMCBNU_6000 finger vein database. This dataset is currently not publicly available, but further information is provided in the following paper "An available database for the research of finger vein recognition":

IEEE paper Page (external link)

SDUMLA-HMT - A Multimodal Biometric Database

The SDUMLA-HMT was established by the Shandong University. It consists of face images from 7 view angles, finger vein images of 6 fingers, gait videos from 6 view angles, iris images from an iris sensor, and fingerprint images acquired with 5 different sensors. The database includes real multimodal data from 106 individuals. The finger vein subset of the SDUMLA-HMT contains palmar finger vein images of left/right index, middle and ring finger from 106 subjects. From all of them 6 fingers per subject and 6 images per finger have been acquired in a single session. This results in a total of 3816 images, each exhibiting a resolution of 320 x 240 pixels. Further information and download:

SDUMLA-HMT Info Page (external link)

CIE - University of Poznan Hand Vein Data Set

This dataset contains 1200 palmar hand vein images captured using reflected light illumination. The images have been acquired for left and right hand in three sessions. All of the images exhibit a resolution of 1280 x 960 pixels. The acquisition system was made up of a low cost USB camera in combination with IR emitting diodes. Further information and download:

CIE information page (external link)

Hand Vein Subset of the PROTECT Multimodal Biometric Database

The hand vein subset of the PROTECT Multimodal Biometric Database contains further two subsets of left and right dorsal hand vein images, acquired from 40 subjects, 5 images per hand. All images exhibit a resolution of 384 \x 384 pixels and were captured using a reflected illuminator equipped with NIR LEDs with a peak wavelength of 950 nm. The capturing device is similar to one used for the PLUSVein-FV3 images, an IDS UI-ML1240-NIR camera with an 950 nm NIR-pass filter. Further information and download:

PROTECT - Dataset page (external link)

The Idiap Research Institute VERA Palmvein Database

The VERA Palmvein dataset contains 2200 palmar hand vein images of both hands captured using reflected light illumination in two different sessions. The images with a resolution of 580 x 680 pixels were acquired by the application of an Imaging Source camera, containing a Sony ICX618 sensor and a NIR illumination of LEDs using a wavelength of 940 nm. Further information:

VERA Palm Vein information page (external link)

Evaluation Framework Information

The experimental evaluations have been conducted using the open source vein recognition framework (PLUS OpenVein Finger- and Hand-Vein Toolkit) provided by the University of Salzburg. This is a feature extraction and matching/evaluation framework for finger- and hand-vein recognition implemented in MATLAB. It was tested on MATLAB 2016 and should work with all version of MATLAB newer or equal to 2016. This software is under the Simplified BSD license.

A more detailed description of the framework as well as its sources can be found here:

The framework contains all the feature extraction, comparison as well as evaluation methods used for the experiments in the paper.

Result Files and Settings

Two resources can be downloaded to obtain the mandatory information necessary for performing the experiments:

• All settings files utilised by the PLUS OpenVein-Toolkit. For the baseline and reconstruction experiments the settings files are the same most datasets. For some the settings differ slightly, thus for baseline and reconstruction separate setting files exist.
• The score and result files obtained during our intra- and inter-dataset evaluation (as presented in the paper). The corresponding baseline results can be re-generated by the use of the datasets, the provided settings files and the OpenVein-Toolkit.