Research in Biometrics Algorithms at the Institute of Information Mathematics, Faculty of Mathematics

 2019.12.3.

The Institute of Information Mathematics, Faculty of Mathematics has organized a research unit to study biometrics which is important in information technology and conducted researches in biometrics algorithms since 2001. Several kinds of biometrics algorithms the institute developed have already reached the highest level in the world.

The institute has been conducting researches into fingerprint recognition, face recognition, iris recognition, finger vein recognition, palm vein recognition and text-independent and text-dependent speaker recognition algorithms.


- Research in Fingerprint Recognition

We have studied fingerprint recognition algorithms since 2004. The focus of this research was to develop algorithms that could be used in fingerprint-based products such as fingerprint attendance devices and fingerprint locks.

In fingerprint recognition research, we investigated the construction of stable minutiae descriptor, fingerprint indexing for fast matching, fingerprint template improvement for improving recognition accuracy and fingerprint preprocessing, thus boosting the performance of our fingerprint recognition algorithm to the world level.

We attained high ranking in palm print contest at FVC-Ongoing in June, 2018 (https://biolab.csr.unibo.it/fvcongoing).


- Research in Face Recognition

We have studied face recognition algorithms since 2001.

The main research direction is algorithm for construction of face recognition system by using near-infrared cameras. Feature extraction and matching methods were studied in order to enhance robustness to eyeglasses, pose variation and distance variation, and improve the accuracy of the system.

The near-infrared face recognition algorithm developed in our institute is now used in various products including attendance devices based on face recognition.

(1)
(2)
Fig. A sample input image to face recognition algorithm and its face detection and fiducial marks extraction result

- Research in Iris Recognition

We have studied iris recognition algorithms since 2015.

We studied methods for iris detection and boundary estimation under various conditions, eliminating impact by eyelids and eyelash, iris image preprocessing and feature extraction in order to improve stability and accuracy of iris recognition.

The iris recognition algorithm we developed ranked high in a series of international contests.

(1)
(2)
(3)
(4)
Fig. Preprocessing results of the iris recognition algorithm: (1) input image, (2) detected pupil, iris and eyelid, (3) normalization result of iris area, (4) preprocessing result.

- Research in Finger Vein Recognition

We have studied finger vein recognition algorithms since 2013.

What is important in increasing accuracy of finger vein recognition is stable detection of finger boundaries in images, construction of geometric transformations and construction of method for reliable detection of vein structures and matching. We made significant successes in solving the above-mentioned problems.

The finger vein recognition algorithm also gained high marks in the international contests and is now widely used in a variety of products.

(1)
(2)
(3)
(4)
Fig. Preprocessing results of the finger vein recognition algorithm: (1) input image, (2) detected upper and lower boundaries of the finger, (3) normalized image of the finger and (4) preprocessing result.

- Research in Palm Vein Recognition

We have studied palm vein recognition algorithms since 2015.

Stable detection of palm under background, illumination and palm pose variation, image preprocessing for detecting exact vein structures, and feature extraction and matching robust to lighting and palm pose changes, are critical in improvement of accuracy of palm vein recognition. We focused on addressing these problems and made a great achievement, and the algorithms we developed are already in use with various products.


(1)
(2)
(3)
(4)
Fig. Processing result of the palm vein recognition algorithm: (1) input image, (2) detected landmarks of palm, (3) normalized image of palm area and (4) preprocessing result.

- Research in Text-independent and Text-dependent Speaker Recognition

We have studied speaker recognition algorithms since 2001.

With the development of mobile communication devices such as mobile phones, research in speaker recognition is becoming critical. We developed high performance text-independent and text-dependent speaker recognition algorithms which are robust to noises and emotion and other variations by combining several features and applying a variety of regularization methods.


We are planning to improve performance of biometrics algorithms we have already studied and conduct researches in biometrics of other modals such as palm print recognition which is needed in practice by combining deep learning, machine learning, pattern recognition and modern image processing technologies with the achievement and experiences so far as the springboard. Furthermore, we are going to launch theoretical and technological research to significantly improve stability and accuracy of multi-modal biometrics which combine various kinds of biometrics technologies as befits world trend.