In recent years, automatic ear recognition has become a popular research. Effective feature extraction is one of the most important steps in Content-based ear image retrieval applications. In this paper, a new approach is proposed that the low frequency sub-images are obtained by utilizing two-dimensional wavelet transform and then the features are extracted by applying Orthogonal Centroid Algorithm to the low frequency sub-images. The experimental results on USTB ear database prove that the proposed method can overcome the Small Sample Size problem and get better performance of recognition speed than conventional PCA+LDA algorithm.