Information theory for gabor feature selection for face recognition

A discriminative and robust featurekernel enhanced informative Gaborfeatureis proposed in this paper for face recognition. Mutual informationis applied to select a set of informative and nonredundant Gabor features,which are then further enhanced by kernel methods for recognition.Compared with one of the top performing methods in the 2004 FaceVerification Competition (FVC2004), our methods demonstrate a clearadvantage over existing methods in accuracy, computationefficiency, and memory cost. The proposed method has been fullytested on the FERET database using the FERET evaluation protocol.Significant improvements on three of the test data sets areobserved. Compared with the classical Gabor wavelet-basedapproaches using a huge number of features, our method requiresless than 4 milliseconds to retrieve a few hundreds of features. Due tothe substantially reduced feature dimension, only 4 seconds arerequired to recognize 200 face images. The paper also unifieddifferent Gabor filter definitions and proposed a training samplegeneration algorithm to reduce the effects caused by unbalancednumber of samples available in different classes.