Iris Identification using Keypoint Descriptors and Geometric Hashing

Mehrotra, Hunny (2010) Iris Identification using Keypoint Descriptors and Geometric Hashing. MTech by Research thesis.



Iris is one of the most reliable biometric trait due to its stability and randomness. Conventional 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 effect. In this thesis, the aforementioned issues are addressed by considering Noise Independent Annular Iris 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 ransformation 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, Harris Corner Points are detected from iris and matched using novel Dual stage approach. Harris corner improves accuracy but fails to achieve scale invariance. Further, Scale Invariant Feature Transform (SIFT) has been applied to annular iris and results are found to be very promising. However, SIFT is computationally expensive for recognition due to higher dimensional descriptor. Thus, a recently evolved keypoint descriptor called Speeded Up Robust Features (SURF) is applied to mark performance improvement in terms of time as well as accuracy. For identification, retrieval time plays a significant role in addition to accuracy. Traditional indexing approaches cannot be applied to biometrics as data are unstructured. In this thesis, two novel approaches has been developed for indexing iris database. In the first approach, Energy Histogram of DCT coefficients is used to form a B-tree. This approach performs well for cooperative databases. In the second approach, indexing is done using Geometric Hashing of SIFT keypoints. The latter indexing approach achieves invariance to similarity transformations, illumination and occlusion and performs with an accuracy of more than 98% for cooperative as well as non-cooperative databases.

Item Type:Thesis (MTech by Research)
Uncontrolled Keywords:Iris,Harris Corner Points,Scale Invariant Feature Transform,Speeded Up Robust Features
Subjects:Engineering and Technology > Computer and Information Science > Image Processing
Divisions: Engineering and Technology > Department of Computer Science
ID Code:2758
Deposited By:Hemanta Biswal
Deposited On:27 Jul 2011 14:35
Last Modified:27 Jul 2011 14:35
Supervisor(s):Majhi, B and Gupta, P

Repository Staff Only: item control page