As a newly emerging biometric technology, iris recognition has been receiving more and more interests. Iris image pre-processing occupies a very important status in iris recognition systems. Iris localization and iris image normalization are the two most important problems in iris image pre-processing. Fast iris localization algorithm can significantly enhance the speed of iris recognition systems; Physiology based iris image normalization method will highly improve the recognition rate. Such two issues are discussed in detail in this thesis. The main contributions of this thesis lie in the following aspects: 1. We improved the traditional RANSAC (Random Sample Consensus) method in two aspects: the choice of the random sample set and the selection of the candidate model. The improved RANSAC algorithm is more efficient with great data volume. The proposed method is a universal parameter estimation algorithm and has wide applications. 2. We applied the improved RANSAC in iris localization and eyelid detection. Experimental results have shown that it could greatly decrease the computational cost without reducing the accuracy. 3. We implemented several commonly used iris localization methods and made a comparison in terms of efficiency, accuracy and influence of recognition rate of an algorithm. Moreover, the trade-off between efficiency and accuracy for each specific method was investigated. 4. A physiology based iris image normalization algorithm was proposed and compared with some conventional methods. Experiments indicated the proposed method could describe the iris motion more reasonably and help to improve the recognition rate.
修改评论