22 May 2007
Centre de Visió per Computador, UAB
Bogdan Raducanu

In real life, visual learning is supposed to be a continuous process. Humans have an innate facility to recognize objects even under less-than-ideal conditions and to build robust representations of them. These representations can be altered with the arrival of new information and thus the model of the world is continuously updated. Inspired by the biological paradigm, we propose in this paper an incremental subspace representation for cognitive vision processes. The proposed approach has been applied to the problem of face recognition. The experiments performed on a custom database show that at the end of incremental learning process the recognition performance achieved converges towards the result obtained using an off-line learning strategy.