A novel borderline preserving embedding manifold learning algorithm

The notorious curse of dimensionality is a well-known phenomenon in pattern recognition. A lot of algorithms have been proposed to find a compact representation of data as well as to facilitate the recognition task. In order to solve the problem of dimension disaster, a novel dimensionality reduction technique called borderline preserving embedding (BPE) is proposed in this paper. Unlike the traditional dimensional reduction algorithms such as principal component analysis (PCA) and linear discriminant analysis (LDA) which project data in a global sense, BPE seeks for a local structure in the manifold. From this perspective, it is similar to other subspace learning techniques. However, BPE has the advantage of preserving the borderline in local reconstruction. Theoretical analysis and experimental study show that the improved manifold learning algorithm can provide better representation in low dimensional space and achieves higher classification accuracy in face recognition in comparison with traditional dimensionality reduction algorithms.