Face recognition using extended isomap

Abstract

The isomap method has demonstrated promising results in finding low dimensional manifolds from samples in high dimensional input space. While conventional subspace methods compute L/sub 1/ or L/sub 2/ metrics to represent distances between samples and apply principal component analysis or similar to induce linear manifolds, the isomap method estimates the geodesic distance between samples and then uses multidimensional scaling to induce a low dimensional manifold. Since the isomap method is based on the reconstruction principle, it may not be optimal from the classification viewpoint. We present an extended isomap method that utilizes the Fisher linear discriminant for pattern classification. Numerous experiments on image data sets show that our extension is more effective than the original isomap method for pattern classification. Furthermore, the extended isomap method shows promising results compared with the best classification methods in the literature.