Biometrics

In personal identification we have two typical problems: "verification" and "recognition". Verification (authentication) refers to the problem of confirming or denying a person's identity: for example, our identity is "verified" when we use a "Bancomat" card. Recognition (identification) refers to the problem of establishing a subject's identity: for example in forensic applications.

The problem of personal identification is very important in today's networked society. However, "traditional" methods, like password, PIN, ATM, are unreliable because a personal code (a sequence of letters or digits) can be stolen or duplicated, and used by other people for illegal aims. In this context, the "biometric" field is a very active research field. Its aim is to find reliable personal identification techniques based on human characteristics like face, fingerprint, retina, signature, iris, gait and so on. A biometric system assures a more reliable identification of a person, since fingerprint, face etc. are unique for each person and cannot be stolen or duplicated.

PRA group is currently working on several open issues about biometrics. In particular:

• Fingerprint classification;
• Fusion of multiple matchers;
• Fingerprint vitality detection, quality evaluation and forensics;
• Template update and selection.

Methods and algorithms are mainly developed and tested for biometric systems based on fingerprints and faces.   

In the following, we give some small details abount these biometrics.

Fingerprints

Fingerprints are the most important among biometrics. Studies about fingerprints started at the end of XIX century with Galton, Gasti and Henry, that published a fundamental book about fingerprint characteristics (E.R.Henry ­ "Classification and Uses of Fingerprints" ­ Routledge, London 1900). Ridges and valleys constitute a fingerprint: ridge-flow determines a particular shape. It is possible to group the set of fingerprint in a certain number of "shapes", named "classes" or "categories". We can identify two points in which the ridge-flow becomes irregular, named "core" and "delta" points. Ridges can be interrupted (ridge ending) or spitted from single paths to two paths (ridge bifurcation). These points are called "minutiae" and are the most important micro-characteristics of a fingerprint: each fingerprint is characterised by a different set of minutiae.

The capability of grouping a fingerprint in classes is very useful in fingerprint recognition: before comparing the candidate fingerprint with all images in a database, we can first identify the class of the fingerprint candidate, and then perform matching only with the images of the identified class. This is justified by the usually large number of fingerprint images in a database. For example, in the FBI database there are more than 70 millions fingerprint images. Hence, "classification" and "matching" are the two main steps in fingerprint recognition.

Faces

Personal identity verification based on face recognition involves three main tasks: face detection in a scene, face identification and face verification. In particular, the distinction between face identification and verification depends on the number of individuals to be matched against. Given a face image, face identification consists in providing the correct label associated with that face from all the individuals in a database. Face verification consists just in deciding if an input face image is associated with a given face image.

The features used to perform face identification and verification are extracted from the grey-level face image. The feature set can be indicated manually (for example by indicating eyes, nose, lips), or can be derived automatically by using connectionist approaches like neural networks and Kohonen maps. Typically used methods are based on PCA ("eigen-faces" ) and LDA ("fisher-faces") techniques. Structural approaches like elastic-bunch graph matching are also used.

People working on this topic:

• Zahid Akhtar
  
• Luca Didaci
• Biagio Freni
• Giorgio Giacinto
• Gian Luca Marcialis
• Ajita Rattani
• Fabio Roli
• Roberto Tronci

Publications on Biometrics