Retour à la page précédente

Detection and recognition of people using the face


Detection and recognition of people using the face is a challenging problem in computer vision, that has many applications such as video-conferencing, multimedia indexing, access control and biometric authentication.

In collaboration with IDIAP, two Windows computer programs have been developed:

  1. Face Tracker: This computer program detects, locates and extracts a face in a video stream. The reliability and response-time of face detection has a major influence on the performance and usability of subsequent processing, such as face recognition. The detector extracts 19x19-pixel sub-windows from a given image. To detect faces at different scales, the original image is progressively downsampled and submitted to the detector [1]. The 19x19 regions are first preprocessed and given to a cascade [2] of coarse-to-fine classifiers. These classifiers are obtained by boosting light invariant features. Finally, arbitration is used to merge multiple detections.
  2. BioLogin and its companion application BLUM: The BioLogin application is a bimodal (audio, image) biometric user verification system. The traditional password used for verification is replaced by a snapshot taken by a webcam. Audio and image features of the snapshot are extracted and compared to a model of the claimed user. This comparison leads to a decision (the person in front of the web cam is or is not the claimed user). The BLUM (BioLogin User Manager) program builds users’ model out of sample snapshots.

Algorithms for face tracking and user verification have been developed by IDIAP. The Industrial System Institute has translated these algorithms into a state of the art object oriented software package for PC. Biologin has been invited by OSEC (Swiss Office for Commercial Expension) to be demonstrated as part of the Swiss Pavillon at the 2006 CeBIT fair in Hannover.