Why to combine reconstructive and discriminative information for incremental subspace learning
Daniel Skocaj, Martina Uray, Aleš Leonardis, Horst Bischof · 2006
In the paper we propose a novel method for in- cremental visual learning by combining reconstructive and discriminative subspace methods. This is achieved by em- bedding LDA learning and classification into the incremen- tal PCA framework. The combined subspace consists of a truncated PCA subspace and a few additional basis vec- tors that encompass the discriminative information, which would be lost by the discarded principal vectors. As such it contains both sufficient reconstructive information to en- able incremental learning, and the previously extracted dis- criminative information to enable efficient classification as well. We demonstrate that we are able to efficiently update the current model with new instances of the already learned classes as well as to introduce new classes.