Discriminative scale invariant feature transform (SIFT FLD) model for efficient representation and accurate recognition of faces

Abstract

In this paper, we propose a new discriminative scale invariant feature transform (SIFTFLD) model for efficient representation and accurate recognition of faces. Discriminative SIFT descriptors are extracted for compact representation of faces. Unlike PCA-SIFT that employs PCA on normalized gradient patch; we employ FLD on smoothed weighted histograms. The proposed model has better recognition performance in an unstructured environment and invariant to in-plane rotations. To establish the superiority of the proposed model, we have experimentally compared the performance of our new algorithm with recently proposed (2D)2-FLD on the benchmark databases: AT&T and CALTECH face datasets. Copyright © 2009 by IICAI.