Binarized statistical features for improved iris and periocular recognition in visible spectrum

Visible spectrum iris verification has drawn substantial attention due to the feasibility, convenience and also accepted performance. This further allows one to perform the iris verification in an unconstrained environment at-a-distance and on the move. The integral part of the visible iris recognition rely on the accurate texture representation algorithm that can effectively capture the uniqueness of the texture even in the challenging conditions like reflection, illumination among others. In this paper, we explore a new scheme for the robust visible iris verification based on Binarized Statistical Image Features (BSIF). The core idea of the BSIF descriptor is to compute the binary code for each pixel by projecting them on the subspace which is learned from natural images using Independent Component Analysis (ICA). Thus, the BSIF is expected to encode the texture features more robustly when compared to contemporary schemes like Local Binary Patterns and its variants. The extensive experiments are carried out on the visible iris dataset captured using both Light field and conventional camera. The proposed feature extraction method is also extended for enhanced periocular recognition. Finally, we also present a comparative analysis with popular state-of-the-art iris recognition scheme.