Posture Invariant Personal Identification System Using Convex Hulls: Ear Biometrics

Abstract

Aims: In this work we have set forth two aims, i. to find a unique methodology to capture the shape of human ears using convex hulls and ii. to develop a rotation invariant personal identification system Study Design: Application of convex hulls to capture the shape of the human ear in a precise manner. Place and Duration of Study: Research Center, Department of Master of Computer Applications, Siddaganga Institute of Technology, Tumakuru, India, between June 2014 and July 2018. Methodology: The work focused in this part is about using convex hulls for capturing the ear shape to utmost accuracy in two different orientations: i. orientation with respect to plane of the ear which accounted for rotation and ii. Orientation with respect to perpendicular axis through the ear plane which accounted for tilting. In order to meet the objective of developing a rotation invariant personal identification system. Thirteen parameters namely area, aspect ratio, bari centric coordinate, convexity, concavity, eccentricity, circular equi-diameter, Euler number, faret’s diameter, form factor, orientation, perimeter and solidity were considered. Results: The system was checked by conducting identification experiments. The recognition rate of 100%, 95%, 85% and 77% was noticed for 00, 22.50, 450, 67.50 orientations respectively when Euclidean distance matching criteria was implemented. Apart from this, similarity measures were also considered for matching test image with template image. In this connection Cosine, Jaccard and Dice similarity measures were used. Cosine similarity measure showed relatively higher recognition rates of 84%, 82%, 75.6% and 74.6% for 00, 22.50, 450, and 67.50 orientations respectively. Similarly Jaccard similarity measure performed with 78%, 75.25%, 74.25% and 72.8% for the four orientations respectively. Dice similarity measure exhibited 75%, 73%, 68% and 72% for the four orientations respectively. The overlapping similarity measure showed a drastic behavior by arriving at only two groups and with reduced recognition rates of 72%, 69%, 67% and 64% respectively. Conclusion: It is concluded that the outcome of the research would be of immense help to the research community in the realm of ear biometrics. In addition, the contribution of head posture invariant person recognition system will definitely inspire the research community as well as the developers of biometric systems to explore the area of ear biometric related personal identification system.