L. Rueda, D. Mery, and J. Kittler (Eds.): CIARP 2007, LNCS 4756, pp. 281–290, 2007.
© Springer-Verlag Berlin Heidelberg 2007
A Method for Segmentation of Local Illumination
Variations and Photometric Normalization in Face
Images
Eduardo Garea Llano, Jose Luís Gil Rodríguez, and Sandro Vega
Advanced Technology Application Center. 7ma, No. 21812, Siboney, Playa, Cuba, 12200
{egarea,jlgil,svega}@cenatav.co.cu
Abstract. In this paper we present a method for the automatic localization of
local light variations and its photometric normalization in face images affected
by different angles of illumination causing the appearance of specular light. The
proposed approach is faster and more efficient that if the same one was carried
out on the whole image through the traditional photometric normalization
methods (homomorphic filtering, anisotropic smoothing, etc.). The process con-
sists in using an algorithm for unsupervised image segmentation based on the
active contour without edges approach with level set representation model for
localization of regions affected by specular reflection combined with a normali-
zation method based on the local normalization that considers the local mean
and variance into the located region. The performance of the proposed approach
is compared through two experimental schemes to measure how the similarity is
affected by illumination changes and how the proposed approach improves the
effect caused by these changes.
Keywords: image segmentation, photometric normalization.
1 Introduction
Face recognition algorithms consist in three major parts: Face detection, normaliza-
tion and face identification [1]. Face recognition starts with the detection of face pat-
terns in sometimes cluttered scenes, continues normalizing the face images to attenu-
ate or eliminate geometrical and illumination problems, then these faces are identified
using appropriated classification algorithms, and finally results are post-processed us-
ing model-based schemes and logistic feedback [2].
One illumination effect that might cause particular problems in the recognition
process is the local reflection of light in the face. Recently many appearance-based
algorithms have been proposed to deal with the problem [3-6]. These algorithms work
well, but are computationally expensive.
To find a method to efficiently and quickly solve these problems that obtains face
images without the specular illumination effect and maintaining the features neces-
sary for identification is a challenge.
In this paper we present a new approach to perform a detection of regions affected
by the specular illumination effect by means of a bi-class unsupervised texture image