A novel approach for ear recognition based on ICA and RBF network

Ear recognition is a new biometrics technique. Due to its unique physiological structure, position and stability, ear recognition is expected to be a promising authentication technique. In this paper, a hybrid system for classifying ear images is proposed. This system combines independent component analysis (ICA) and RBF network. The original ear image database is decomposed into linear combinations of several basic images. Then the corresponding coefficients of these combinations are fed up into RBF network instead of an original feature vector comprised of pixel values of grayscale images. The local features extraction of ICA and the adaptability of RBF neural network are combined reasonably. The robustness of the system is enhanced. The experiment results show that the recognition rate of ICA RBF method is improved substantially.