We explore the various problems related to the analysis of the presence of humans in images and image sequences.
Starting off from our face detection procedure , our current research mainly focuses on the two following applications:
- Identity validation systems for access control, where we deal with the problem of validating identities (see Fig. 1) by means of supervised learning techniques;
Fig. 1: the comparison between a recognition (above) and verification (below) based application.
- Real time face recognition for visually impaired people , in which we consider an unsupervised setting and exploit the combination of face matching techniques and identity verification (see Fig. 2) to achieve a recognition rate that competes with the one-vs-all classification but requires less computational resources. We also deal with the face recognition problem from a semi-supervised perspective, which is very appropriate since the set of user's everyday observations typically includes only a (relative small) subset of recognized faces -- i.e. labeled examples -- as opposed to an higher number of unlabeled data.
Fig. 2: the pipeline of face recognition that combines face matching with identity verification.
Inspired by the specific requirements of the above-mentioned applications we try to exploit the availability of spatio-temporal information and study their applications in different learning frameworks.
- Imavis (R&D),
- Istituto Chiossone (Usability, accessibility, testing for visually impaired users),
- DICO Università di Milano (DBMS)
- Destrero, A. et al. "A Regularized Framework for Feature Selection in Face Detection and Authentication". IJCV (2009).
- Luigi, Balduzzi et al. "Low-cost face biometry for visually impaired users". 2010 IEEE Workshop on Biometric Measurements and Systems for Security and Medical Applications, 2010.