Ear Recognition and Occlusion

Personal identification using 2D ear images still has many problems such as occlusion mostly caused byhair, earrings, and clothes. To avoid this problem, we propose to divide the ear image into non-overlappingequal divisions and identify persons through these non-occluded parts separately and then combine outputsof the classification of these parts in abstract, rank, and measurement level fusion. Experimental resultsshow that the increasing of recognition rate through combining small parts of non-occluded divisions ofear image.