Modifications on illumination, distance function and Gabor masks for elastic bunch graph matching

Abstract

Face recognition is one of the most important field in computer vision, although there are many proposals and research papers still have limitations in real applications where uncontrolled conditions such as illumination, view angle, facial expressions, resolution and image quality, etc. are the main problems. To solve these issues there are a large number of methods for recognition, which can be grouped according to the approach by which the recognition process is addressed, such groups are; holistic methods and featrure based. In this paper we address a feature-based method called Elastic Bunch Graph Matching (EBGM) which is appropriate for an uncontrolled environment for its tolerance to changing background and certain variations of pose, also by having a lower sensitivity variations of illumination which is one of the weaknesses in the holistic methods, which compared to them EBGM without any modification obtained results alongside well-known methods as Principal Component Analysis (PCA). EBGM has several parameters that can be configured and most of the works in literature uses the default settings of the original author. In this respect this paper presents modifications to these parameters as regards the number of models, illumination enhancement in the preprocessing phase, the configurations Gabor masks and modifying the similarity function. Finally we corroborate the existence of improvements on our experimental results.