Multispectral iris recognition using patch based game theory

Abstract

Multispectral imaging offers potential to improve the recognition performance of an iris biometric system. The novelty of this research effort is that a Coalition Game Theory (CGT) is proposed to select only the important patches that are obtained using the modified Local Binary Pattern (mLBP) operator. The mLBP fuses both the sign and magnitude difference vector in an effort to extract feature from normalized iris images. The CGT selects patches based on the Shapely value that have better individual importance along with a strong interaction with other patches to improve the overall performance. Results show that CGT model maintains better recognition accuracy while reducing the overall surface area needed for recognition purpose.