Automatic Ear Classification System using Lobule Structural Shape

Abstract:  The rate at which ear recognition system is been explored by computer vision authors and its acceptability as a biometric traits has increased. Ear classification scheme has become necessary due to anticipated exponential growth of the ear database which would run to millions in the nearest future. Thus, in this paper, we proposed automatic ear classification scheme so that the ear can be classified into two distinct groups; lobed ear and lobeless ear using geometric structure of lobule. The ear image is cropped and its contrast normalized to produce more quality image. Ear contour image is localized since it can reveal the shape of the ear image. Due to the rumpling nature of the ear, the ear contour harbours many contour images, some which are noisy contours. The ear contour image is enhanced to remove the outer ear contour from the ear contour for feature selection. Using chain code element, ear outer shape is encoded and lobule code signature extracted. Forming a lobule feature code bag, a threshold for the class partition of the ear is established. Experimental result is carried out with University of Science and Technology Beijing (USTB) ear image, which is divided into three samples; 77, 154 and 308 samples respectively. The proposed method achieved an ear classification accuracy of 96.1%, 92.20% and 88.96% under these respective samples.