Lighting normalisation for face recognition

A method is presented that detects lighting variations within a face image, and transform the image into a front-lit image. It contains two modules: lighting-effects determination and lighting-effects normalisation. In the first module, lighting effects are determined using the multivariate discriminant analysis and the wavelet packets transform. In the second module, the determined lighting effects are normalised by rendering a front-lit face image using the linear object classes and the principle components analysis. The performance of the method is evaluated and the results are reported. © 2005 IEEE.