| Summary: | 碩士 === 國立暨南國際大學 === 電機工程學系 === 94 === Through the whole human life, the structure of iris does not change unless the eyes get diseased. That is its advantage as well as disadvantage. Advantage: the Iris is highly stable and can result in very good recognition. Disadvantage: there will be a great loss if it is stolen or counterfeited. We will study deeply the methods for recognition and anti-forgery from iris image sequence. The system architecture includes five modules: image acquisition, detection liveness, image pre-processing, feature extraction and recognition modules. Firstly, our system captures a sequence of iris images of a human eye. The detection liveness module adopts coefficients from the input iris sequences to determine if the eyes are live or just replay the fake iris images. The preprocessing module uses some image processing algorithms to extract the region of interest of iris from the sequence of human eye images, it can use differences resulted from the iris image edge detection and the area of light reflection to estimate the iris image quality. Then, the extraction module uses Laplacian pyramid transformation and eigeniris to extract the characteristic codes for identity. Finally, the system applies these characteristic codes for iris matching in recognition module.
We have designed experiments for some assumed interesting scenarios, and launched them on our video database which consists of 175 iris image sequences of 25 classes, and 175 fake iris image sequences. The proposed mechanism can successfully detect the forgery attacks. The differentiation forgery rate of the system for video input data can achieve up to EER=96.57%. Even under the circumstance of false acceptance rate （FAR）0%, the system still approaches the differentiation rate in 92.43%. This thesis analyzes the experiment results and verity the proposed methods for further research.
|
|---|
