Spatially Adaptive Logarithmic Total Variation Model for Varying Light Face Recognition

Biao WANG  Weifeng LI  Zhimin LI  Qingmin LIAO  

IEICE TRANSACTIONS on Information and Systems   Vol.E96-D   No.1   pp.155-158
Publication Date: 2013/01/01
Online ISSN: 1745-1361
DOI: 10.1587/transinf.E96.D.155
Print ISSN: 0916-8532
Type of Manuscript: LETTER
Category: Image Recognition, Computer Vision
face recognition,  illumination normalization,  logarithmic total variation (LTV) model,  

Full Text: PDF>>
Buy this Article

In this letter, we propose an extension to the classical logarithmic total variation (LTV) model for face recognition under variant illumination conditions. LTV treats all facial areas with the same regularization parameters, which inevitably results in the loss of useful facial details and is harmful for recognition tasks. To address this problem, we propose to assign the regularization parameters which balance the large-scale (illumination) and small-scale (reflectance) components in a spatially adaptive scheme. Face recognition experiments on both Extended Yale B and the large-scale FERET databases demonstrate the effectiveness of the proposed method.