Less Data for Face and Iris Biometric Traits: An Approximation for Feature Extraction

Abstract

Biometrics applications have been emerging since the evaluations on the sensors. From fingerprint to iris of the human being, several biometric traits have been used in many security applications. Biometric e-passports use embedded biometric data for person identification. As the data consumption rate increases, a huge amount of data processing need occurs. In the 2020s, it is forecasted that a person will be consuming data almost half of terabytes per day. This is the motivation of approximations in data which helps to reduce the computation load and data storage overhead. In this research, a previously proposed pixel value comparison based operator within a window in image processing namely Relational Bit Operator (RBO) will be revisited by measuring its data approximation property. The less data versus accuracy tradeoff over biometric traits of face and iris is going to be tested. Having used both of the face and iris datasets allows us to see how much it deviates from ideal classification results when there is used the approximate biometric features. Moreover, the proposed method has an underlying motivation on the data occurrence, thus an ease of use for data reduction is emphasized. Therefore, the purpose of this paper is to build a clear understanding of biometrics to have the methods that treat limited data storage. Following sections start with the introduction, continues with motivation, methods, system details, tests and end up with a conclusion.