Comparative analysis of hashing schemes for Iris identification using local features

Abstract

Iris is one of the most reliable biometric trait due to its stability and randomness. Traditional recognition systems transform the iris to polar coordinates and perform well for co-operative databases. However, the problem aggravates to manifold for recognizing non-cooperative irises. In addition, the transformation of iris to polar domain introduces aliasing eect. In this thesis, Noise Independent Annular Iris is used for feature extraction. Global feature extraction approaches are rendered as
unsuitable for annular iris due to change in scale as they could not achieve invariance to transformation and illumination. On the contrary, local features are invariant to image scaling, rotation, and partially invariant to change in illumination and viewpoint. To extract local features, Scale Invariant Feature Transform (SIFT) has been applied to annular iris. However, SIFT is computationally expensive for recognition due to higher dimensional descriptor. Thus, a keypoint descriptor called Speeded Up Robust Features (SURF) is applied to mark performance improvement in terms of time as well as accuracy. At last, a recently developed Binary Robust Invariant Scalable Keypoints (BRISK) is applied. BRISK performs at a dramatically lower computational cost than SIFT and SURF. For identication, retrieval time plays a signicant role in addition to accuracy. Traditional indexing approaches cannot be applied to biometrics as data are un- structured. In this thesis, two novel approaches has been applied for indexing iris database. In the rst approach, indexing is done using Geometric Hashing of local feature keypoints. This approach achieves invariance to similarity transformations, illumination, and occlusion and performs with a good accuracy for cooperative as well as non-cooperative databases, but it takes larger time for recognition. In the second approach, enhanced geometric hashing is applied using local keypoint descrip- tors of annular iris for dierent databases. Comparative analysis shows that enhanced geometric hashing is more accurate and faster than traditional geometric hashing.