A Mutual Information Based Face Recognition Method

A mutual information based method for face recognition has been proposed. By comparing the mutual information of images locally, this method becomes robust to illumination variation. The method's performance has been evaluated using AT&T database with different number of samples in the training set, as well as different resolutions for intensity distribution estimation. The accuracy rate is dependent on the number of samples in the training set and the accuracy of the probability density function (PDF) estimation. An accuracy rate of 94.58% has been obtained when half of the database was used as the training set and the PDFs were estimated with 20-bin histograms. A perfect accuracy rate was achieved when 60% of the database was allocated to the training set.