Incremental null Foley-Sammon transform
The Foley-Sammon transform (FST) is one of the most well-known dimensionality reduction and feature extraction methods. However, the classical FST cannot be used directly in the small sample size problem where the within-class scatter matrix is singular. Null-space based FST (NFST) provides a good solution to this problem. Proposed is a fast incremental NFST (INFST). INFST extracts new information brought by newly-added samples and integrates it with the existing model by an efficient updating scheme. INFST can achieve the aims of online classification and novelty detection. Experiments on real-world data demonstrate the effectiveness of INFST.