Illumination normalisation method using Kolmogorov-Nagumo-based statistics for face recognition

Illumination compensation has proven to be crucial in many machine vision applications including face recognition. This is especially important in non-controlled scenarios where face illumination is not homogeneous. An extension of the local normalisation (LN) method using Kolmogorov-Nagumo-based statistics to improve face recognition is proposed. The proposed method is a more general framework for illumination normalisation and it is shown that LN is a particular case of this framework. The proposed method using two different classifiers, PCA and local matching Gabor, on the standard face databases Extended Yale B, AR Face and Gray FERET is assessed. The method reached significantly better results than those previously published on the same databases.