Ear Recognition using a novel Feature Extraction Approach

Most of traditional ear recognition methods that based on local features always need accurate images alignment, which may severely affect the performance. In this paper, we investigate a novel approach for ear recognition based on Polar Sine Transform (PST); PST is free of images alignment. First, we divide the ear images into overlapping blocks. After that, we compute PST coefficients that are employed to extract invariant features for each block. Second, we accumulate these features for only one feature vector to represent ear image. Third, we use Support Vector Machine (SVM) for ear recognition. To validate the proposed approach, experiments are performed on USTB database and results show that our approach is superior to previous works.