The algorithm of descriptor based on LPP and SIFT

This paper presents a novel algorithm to design the descriptor (LPP-SIFT) of image feature points based on Locality Preserving Projections (LPP) and Scale Invariant Feature Transform (SIFT). Firstly, the high dimensional feature vectors are extracted by SIFT, which are composed of the gradient vectors of all neighborhood points around feature points. Secondly, the high-dimensional gradient vector is embedded into a low-dimensional manifold space with the eigen matrix by LPP, then a low-dimensional descriptor of the feature points is achieved. The proposed algorithm takes into account neighborhood information of feature vectors, preserves invariability on the geometric structure and keeps good distinguish between feature points. Experimental results suggest that the proposed algorithm provides low-dimensional descriptors for saving memory, speeding up feature matching and improving matching accuracy.