Why to combine reconstructive and discriminative information for incremental subspace learning

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