Unconstrained iris recognition using F-SIFT

Abstract

This paper proposes an efficient approach known as F-SIFT, for unconstrained iris recognition. The acquired iris image is preprocessed to find the annular region underlying inner and outer iris boundary. From the annular iris image, keypoints are detected using traditional SIFT approach. Local features are defined for a window around every detected keypoint using Fourier transform. Finally, the keypoint descriptors are paired using Phase-Only Correlation. From the experimental results it is inferred that F-SIFT performs better in comparison to traditional SIFT approach on publicly available BATH and CASIAV3 iris databases.