Iris Recognition Research Based On Filtering Property

With the development of the network, communication and information technology, identity recognition becomes an important and universal problem. Because of the property of stability, reliability and uniqueness, biometric technology is becoming an active topic. Iris recognition, as one of the biometric technologies, has higher stability, higher reliability and higher security, that it has wide development prospects, important practical significances and application values, and has attracted more and more attentions.Iris recognition system is generally composed of iris image capturing, image preprocessing, feature extraction and matching. Image capturing is the capture of original eye image including iris by capturing device. Image preprocessing includes localization, normalization and denoising. Feature extraction is to extract features benefited for recognition from the images after preprocessed. Matching is comparing and classifying the features with some similarity measures. Where, feature extraction is the kernel of the iris recognition system. The main work of our research is proposing some iris recognition methods based on filtering property, in which we use empirical mode decomposition, FIR filter and Zernike moments to extract features with certain frequency of images, and select corresponding preprocessing and matching methods for recognition.Empirical mode decomposition, which is a non-stationary, nonlinear signal analysis method, can decompose a signal into a series of stationary and linear intrinsic mode functions. This thesis proposes an adaptive filtering iris recognition method based on empirical mode decomposition. With the time-space filtering property, iris signal is decomposed and filtered, and optimal intrinsic mode functions are selected and combined as the most useful features. The downsampling algorithm related to the empirical mode decomposition is used for features compression, and Hamming distance is used as similarity measure for recognition. FIR filter is the finite impulse response filter, which has linear phase property and stability. This thesis proposes an iris recognition method with fix parameters filtering based on FIR filter. Optimal filter, whose bandpass frequency and order are determined by the distance separability criterion, is used to filter the iris signals to extract features whose dimension is further reduced by downsampling algorithm to save memory space and improve recognition speed.Zernike moments, as one type of the image moments, can extract features with rotation invariance of the images. Zernike moments also have the filtering property, for its low order moments represent the low-frequency characteristic while the high order represent the high-frequency information of the images. This thesis proposes an iris recognition method based on Zernike moments and ring normalization. First, the iris images are normalized into a ring and Zernike moments is used to extract rotation invariant features. Then the orders of Zernike moments are selected and the features with certain frequency are determined. Many similarity measures, such as Euclidean distance, Cosine similarity and support vector machine, are adopted for matching. The features extracted by Zernike moments have global characteristic and less amounts, but the images require high quality, that some problems still exist.With the experimental simulation of the above three methods based on CASIA1 database, results are analyzed and the performances are evaluated separately. In the end, the future works and research are prospected.