An efficient iris recognition using local feature descriptor

This paper presents a robust iris recognition system using local feature descriptor. The proposed biometric system accounts for two crucial issues. Firstly, iris texture is usually occluded by upper and lower eyelids. To handle this problem, a novel sector based normalisation is proposed. In this approach only non-occluded region is extracted by forming sectors of variable size. Secondly, texture features of iris transforms linearly due to illumination and position of these features changes due to rotation. For this purpose speeded up robust features (SURF) are found to be useful and invariant to transformations. The system is rigorously tested on database collected from three different sources i.e., BATH, CASIAV3 and IITK. Several local and global approaches have been compared with SURF. Experiments show that SURF outperforms other existing approaches in terms of accuracy and speed.