Gabor wavelets and General Discriminant analysis for ear recognition

Gabor filter is a powerful tool for texture analysis, and (lie Gabor filter representations of images are robust for illuminations, pose and expressional variability, Gabor filters are now being used extensively and successfully in various computer vision applications- However, tbe dimensionality of the Gabor feature space is overwhelmingly high, many sampling or compressing methods are proposed to reduce the space dimension to avoid dealing with the enormous data. In this paper, a novel and uniform framwork for ear recognition is proposed. We describe an approach that combines Gabor feature extraction and General Discriminant Analysis techniques to produce a robust system framework for ear recognition. It not only effectively reduces the dimension of feature vector, but also involves extracting discriminant features to be propitious to classify. In addition, the design of optimized Gabor filter is also discussed based on experiments. Extensive experiments have been conducted to evaluate the performance. Our method has achieved 99.1% recognition rate on the USTB database, and the total time of feature extraction and matching is Q.56S9s, so it meets the requirements of a real-time biometrics system.