Adaptive subspace incremental PCA based online learning for object classification and recognition

The learning method for visual object recognition that compute a space of eigenvectors by Principal Component Analysis(PCA) traditionally require a batch computation step, in which the only way to update the subspace is to rebuild the subspace by the scratch when it comes to new samples. In this paper, we introduce a new approach to object recognition based on online PCA algorithm with adaptive subspace, which allows for complete incremental learning. We propose to use different subspace updating strategy for new sample according to the degree of difference between new sample and learned sample, which can improve the adaptability in different situations, and also reduce the time of calculation and storage space. The experimental results show that the proposed method can recognize the unknown object, realizing online object knowledge accumulation and updating, and improving the recognition performance of system.