Face recognition with Multilevel B-Splines and Support Vector Machines

Abstract

This paper presents a new face recognition system, based on Multilevel B-splines and Support Vector Machines. The idea is to consider face images as heightfields, in which the height relative to each pixel is given by the corresponding gray level. Such heightfields are approximated using Multilevel B-Splines, and the coefficients of approximation are used as features for the classification process, which is performed using Support Vector Machines. The proposed approach was thoroughly tested, using ORL, Yale, Stirling and Bern face databases. The obtained results are very encouraging, outperforming traditional methods like eigenface, elastic matching or neural-networks based recognition systems.