An improved LPP algorithm

Discriminant Locality Preserving Projection (DLPP) has been successfully used as a dimensionality reduction technique to many classification problems, which incorporate discriminant information into Locality Preserving Projection (LPP) to improve recognition rate. However, in order to avoid small sample size problem, DLPP needs to reduce dimensions, which will lose some important discriminative information. Direct Linear Discriminant Analysis (DLDA) can solve the problem by diagonalization. Inspired by DLDA, we propose a novel method of improvement algorithm, which incorporate DLDA into LPP. Compared with DLPP and LPP, this algorithm not only preserves more effective discriminative information, but also solves the small sample size problem in dimensionality reduction. It also improves light sensitivity when distinguish an uneven illumination image. The modified LPP algorithm achieves better result than DLPP and LPP in face recognition.