Cluster Based Ear Biometric System for Personal Identification: An Exploitative Approach

It is becoming increasingly clear among biometric research fraternity that Ear as a biometric articulation in human beings provides exclusive and unique advantages when compared with other kinds. In this paper, we present a person identification system which is based on clustering of ears. For the development of the system, a database of 605 ear images was considered. Shape based biometric features that are related to moment of inertia (MI) of the ear such as MI about major axis, MI about minor axis, radii of gyration with reference to major and minor axis and the area of the ear were elicited presuming that the ear as a planar surface of irregular shape. Extensive computational clustering experiments were done involving kmeans, back propagation network (BPN), four similarity metrics (Jaccard, Dice, Cosine, and Overlapping), radial basis function network (RBFN), and support vector machine (SVM). The results showed that Jaccard, Dice similarity measures and RBNF showed excellent performance in terms of minimal overlapping among groups followed by recognition accuracy ranging from 94%-98%. Further, there was a decrease in CPU time for image matching and retrieval of personal details to an extent of 14 % when compared with the time expended for the same task when the image database was presented in an unorganized way. Keywords— Ear biometrics, shape related parameters, moment of inertia, radius of gyration, major axis, minor axis, similarity measures.