Illumination normalization using local variance in logarithmic domain

Abstract

It is very challenging to obtain the high recognition rate on face recognition under varying illumination because face appearance is changing remarkably by the lighting variation. The lighting variation can be deal as a low frequency multiplicative component according to the Lambertian reflectance model. In this project, we come out with the method using local variance in the logarithmic domain for the illumination normalization. Our proposed methods fLVF and aLVF can achieve a very high recognition rate in the experimental results using the Matlab software. By slightly contribute the accuracy on recognition rate, the proposed methods of fLVF and aLVF would be more competitive in the real-time application.