Face Recognition Based on Unit-Linking PCNN Time Signature

In this paper, a novel method is proposed for face recognition based on pulse coupled neural network (PCNN) time signature. In this approach, a probe face is first extracted PCNN time signature as the recognition features, which a two-dimensional image is projected to a low one-dimensional feature space and then is classified based on the known samples. An extensive experimental investigation is conducted using AT&T face database covering face recognition under controlled/ideal conditions, different illumination conditions and different facial expressions. The recognition ratio is 83.75% when only PCNN time signature, including 5 features is used as the recognition features and three exemplar images per person are available. The recognition ratio is 94.50% when the PCNN time signature is integrated with PCA feature based on Euclidean distance. The recognition ratio is 98.50% when the PCNN time signature is integrated with ICA feature based on support vector machine (SVM). The simulation result demonstrates that proposed recognition method is robust due to the translation, rotation and scale invariance of PCNN time signature, insensitive to illumination and facial expressions changes as well as fast due to the low feature dimensions.