A New Orthogonalization of Locality Preserving Projection and Applications

Locality Preserving Projection (LPP) is a linear projection method that preserves the local structure to find the underlying manifold of the data. Non-orthogonality of LPP basis makes its use difficult not only for reconstruction but also for other applications such as denoising. At present, orthogonal basis of LPP (OLPP) are obtained in an iterative manner which is computationally expensive. In this article, a new orthogonalization of LPP (NOLPP) basis is proposed by relaxing the constraint used to minimize the objective function giving rise to the basis. The reducibility capacity of NOLPP for data clustering is validated by performing experiments on several databases. Use of NOLPP for image denoising shows its efficiency in comparison to the state of the art research. Fine structures present in the images are preserved even at high noise levels.