Occlusion detection and index-based ear recognition
Madeena Sultana, Padma Polash Paul, Marina L. Gavrilova · Digital Library (University of West Bohemia) · 2015
Person recognition using ear biometric has received significant interest in recent years due to its highly discriminative nature, permanence over time, non-intrusiveness, and easy acquisition process. However, in a realworld scenario, ear image is often partially or fully occluded by hair, earrings, headphones, scarf, and other objects. Moreover, such occlusions may occur during identification process resulting in a dramatic decline of the recognition performance. Therefore, a reliable ear recognition system should be equipped with an automated detection of the presence of occlusions in order to avoid miss-classifications. In this paper, we proposed an efficient ear recognition approach, which is capable of detecting the presence of occlusions and recognizing partial ear samples by adaptively selecting appropriate features indices. The proposed method has been evaluated on a large publicly available database containing wide variations of real occlusions. The experimental results confirm that the prior detection of occlusion and the novel selection procedure for feature indices significantly improve the biometric system recognition accuracy.