A Face Recognition under Varying Illumination - Advances in Image and Graphics Technologies - page 120

A Face Recognition under Varying Illumination

Haodong Song 1,2 , Xiaozhu Lin 1 , and Zhanlin Liu 1,2

1 School of Information Engineer, Beijing Institute of Petro-chemical Technology,
Beijing 102617, China
{hd_biptgra,linxiaozhu,zhanlinliu}@bipt.edu.cn
2 College of Information Science and Technology,
Beijing University of Chemical Technology, Beijing 100029, China

Abstract. Face recognition is one of the focus studies in biometrics technology.
The recognition accuracy always changes drastically in different environment,
especially when it is affected by the illumination. Retinex is a method which
utilizes illumination invariant, but it ignores contributions of low frequency
component to face recognition. In this research paper, we propose a face recog-
nition method based on retinex and wavelet transformmation. First, illumination
invariant and variant are generated by the retinex theory. Second, decompose the
illumination component via wavelet transformation and set its low-frequency
coefficients to zero. In doing so, the processed illumination component is
obtained by inversing the transformation. In the end, a new image is acquired
by restructuring the two components. The recognition experiment will demon-
strate that the proposed method ensures good performance in illumination
environment.

Keywords: Face recognition, Retinex, illumination invariant and wavelet
transformation.

1 Introduction

Face recognition is a challenging field in pattern recognition. It is difficult to satisfy the
practice application because of the low recognition accuracy under illumination, pose
and facial expression. In order to solve the problem of illumination, researchers pro-
posed a variety of approaches, such as illumination normalization [1,2], 3D illumina-
tion model [3,4] and illumination invariant [5,6]. The illumination normalization is a
method based on image gray transformation. Though it can be easily implemented, the
result doesn't satisfy the demand under complex illumination. The illumination model,
a way establish face model by different images under different illumination conditions,
achieves a good accuracy in laboratory. However, it's impossible to obtain a plenty of
images in real time face recognition system. For the last category, Land [6] proposed a
retinex theory which suggests that the color of the object depends on the capability of
the light wave reflection and is consistent without the effect of light discontinuity.