Biometrics is the science that seeks to identify a person through physiological characteristics (fingerprint, face) or behavioral (signature, step), "measurable" by appropriate algorithms, called the "biometric". These features are unique from person to person and have a big advantage compared to traditional methods of identification (e.g. password or PIN) can not be forgotten or stolen and with much difficulty may be reproduced.

The main research and development laboratory Ambient Intelligence with regard to biometric methods for identity verification are:

1. Methods for classification of fingerprints

The study of fingerprints has taken hold in the two decades since the personal identification through this method is very safe and reliable. In general, the issues covered within the classification are two types of classification and comparison. In the first case to involve an imprint of the five fundamental classes established by Edward Henry (1900), certain forms of furrows and ridges; The second aspect is to make a comparison between the fingerprint to identify the fingerprints of a certain class where there are samples in a database.

2. Fusion system for checking fingerprints

Many electronic devices have recently been developed to acquire digital images of fingerprints. The methods used for the representation of fingerprints suffer disadvantages of each of the digital image processing capability, whether it is based on "minutiae" that is trying to describe the flow through the imprint of grooves and ridges skin. Consequently, even algorithms direct comparison between two fingerprints, traces of such limits. In order to overcome the limitations of each system was proposed "merger" of more systems, each based on different representations footprint.

Various studies conducted in the Department of the University of Cagliari DIEE showed very promising results.

3. Fusion algorithm for the recognition and verification of faces

The face is the primary means by which people are recognized. So it is universally accepted as a useful tool for automatic recognition of identity. Many methods have been proposed for both identity verification applications (for replacement identification card or password) and for identification of suspects in controlled environments such as airports and general surveillance applications. The critical aspect of these methods is their poor tolerance to environmental changes (brightness, background, distance from the camera) and characteristics of the subject picture (hair, beard, expression), in addition to the difficulty of locating the face in a complex scene. To try to increase flexibility and overall performance of these methods, but not increase the cost and complex implementation of the system, has proposed a blend based on statistical representations of the face. These methods are commonly called PCA and LDA. The results were interesting: on the one hand, have highlighted the increased tolerance to environmental changes, pose and expression, especially the other, showed the ability to exploit the characteristics of individual methods, in unfavorable conditions, allowing performance always best in the worst of the two methods, and in some cases, similar to the better of the two methods. Situations in a controlled environment, the performances were superior to the best method on its own.

4. Recognition of fake fingerprints

The increasing spread of commercial systems based on fingerprints, designed to access control, determined the need to protect themselves have attempted attacks derived from products with false stamps or silicone gel, reproducing fingerprints of persons authorized to access the system . The laboratory environment shares the know-how on this sensitive area with the department DIEE that recently proposed methods to detect if an image of a fingerprint acquired by electronic sensor, is a fake. Preliminary experiments have delivered encouraging results in competition with the most relevant state of the art work.