Foto del docente

Annalisa Franco

Associate Professor

Department of Computer Science and Engineering

Academic discipline: ING-INF/05 Information Processing Systems


Keywords: Face recognition Biometric systems Human action recognition

  • Biometric systems
  • Human activity recognition
  • Image databases

Biometric systems

A biometric system is an automated method of identifying an individual based on his/her physiological (fingerprint, face, hand, retina, iris, ...) or behavioral (voice, handwriting, keystroke style, ...) characteristics. Many different biometric characteristics can be exploited, and each biometric trait presents peculiar characteristics in terms of individuality, acquisition procedure and degree of acceptance. In the last few years, the interest in biometric systems has grown significantly: academic research groups and companies have spent a lot of resources to study highly accurate personal identification systems. In fact, automated identification systems can be very useful in several applications (e.g., physical access control, time and attendance, automatic surveillance, network security, online transactions, …).

Among the possible biometric characteristics, face represents one of the more interesting since the acquisition procedure in not invasive (it does not need to use particular devices or scanners such as those adopted for fingerprint acquisition) and does not require the user cooperation. Although the recognition accuracy of this biometric is not as high as others (e.g. fingerprints), there are many fields where face recognition find its ideal application. An example is video-surveillance in environments where there is a high risk of terrorist acts (airports, stations, embassies): the use of a face recognition system could raise timely alarms when dangerous individuals are identified; analogously, face-based anti-robbery protection systems could be installed in banks and post offices, to alert about the presence of known criminals. In some applications, face is the only biometric characteristic that can be employed to guarantee that the user of a given system (e.g. a computer or a cashier's window) continues, during the whole work session, to be the same who initially logged in. Moreover, face has been chosen as the primary biometric feature for identity verification in electronic ID documents.

The current research activity is mainly focused on face recognition applied to electronic ID documents. More specifically on the following topics:

  • Face image quality assessment and image ISO/ICAO compliance;
  • Analysis of vulnerability of face recognition systems to digital image manipulations;
  • Face morphing attack detection.

Human activity recognition

Automatic recognition of human activities plays a fundamental role in the design of intelligent solutions for home environments, in particular in relation to assisted living applications, where the support of an automated system could improve the quality of life of people by recognizing unusual behaviors or promptly reporting dangerous situations (e.g. falls). In this field of application, the use of RGB-D sensors is widespread, capable of simultaneously acquiring RGB images and depth information that often offer complementary characteristics for recognition.

The current research activity focuses on the design of multimodal approaches for human activity recognition and gesture recognition based on the analysis of RGB images and skeleton information acquired through RGB-D sensor (e.g. Kinect). Semi-supervised template updating techniques are also designed to continuously update the initial templates thus exploit the availability of heterogeneous sources to learn new information from the available continuous data stream.

Image databases

An image database (IDB) is a system in which a large amount of image data and their related i nformation are stored and managed.

Major objectives of an IDB are efficient storage, powerful retrieval capability, flexible data manipulation, efficient indexing. The research in this field is focused on two main issues: efficacy of similarity searches and efficiency in data retrieval in image databases.

As to similarity searches in image databases, some relevance feedback techniques are studied to improve the retrieval results. Relevance feedback approaches are based on a continuous interaction with the user that, providing a judgment on the result obtained for a query, allows the system to gradually adapt its similarity metric to better capture the user perception.

The research activity is also focused on the study of indexing techniques for multidimensional data. In particular indexing techniques have been designed specifically to deal with high dimensional data that present peculiar characteristics that make them very difficult to manage.

Latest news

At the moment no news are available.