Extended Locally Linear Embedding with Gabor Wavelets for Face Recognition

Abstract

Many current face recognition algorithms are based on face representations found by unsupervised statistical methods. One of the fundamental problems of face recognition is dimensionality reduction. Principal component analysis is a well-known linear method for reducing dimension. Recently, locally linear embedding (LLE) is proposed as an unsupervised procedure for mapping higher-dimensional data nonlinearly to a lower-dimensional space. This method, when combined with fisher linear discriminant models, is called extended LLE (ELLE) in this paper. Furthermore, the ELLE yields good classification results in the experiments. Also, we apply the Gabor wavelets as a pre-processing method which contributes a lot to the final results because it deals with the detailed signal of an image and is robust to light variation. Numerous experiments on ORL and AR face data sets have shown that our algorithm is more effective than the original LLE and is insensitive to light variation.