Average Local Binary Pattern Based Iris Recognition

Biometric technology has great significance and wild application prospect in government, business and personal applications. Iris is considered as a competitive biological characteristic because of its high degree of distinction, stability, non-invasion and hardness to counterfeit. Iris Based biometric technology is considered as the most reliable and potential biometric technology in the future.First of all, the thesis introduces the concept of identification recognition and iris recognition. Then we emphasize the advantage by using iris as biological characteristic and the framework of iris recognition. Furthermore, we introduce the classical algorithms of positioning of iris pictures and tell the details about the algorithm we use in the thesis to position and segment the iris. Finally, we study the LBP methods for the extraction of iris characteristics and validate their effectiveness from different kinds of experiments.In the positioning part, based on the gray distribution characteristic of the eye pictures, we use a method based on choosing three points on the iris. The method can be divided into four steps:find one point in the pupil, calculate the gradient, find the boundary of pupil and iris (inner boundary), find the boundary of sclera and iris (outer boundary). The segmentation and normalization part introduces the method of normalization and how to reduce the influence of eyelid and eyelash. The positioning method used in this paper is easy for calculation, with high speed and accurate to find the iris region.In the characteristic extraction part, we focus on the LBP method and proposed an ALBP operator. Based on that, we proposed an iris characteristic extraction algorithm with lower computation complexity, which means a fast speed of calculation. Such algorithm is suitable for applications with higher demand of real-time system.In the experiment part, we use three databases. They are CASIA-IrisV4-IntervaK CASIA-Iris-Thousand and UBIRIS.V1. We introduce the databases and the way we choose the iris images from them. For the classification part, we use the NN and SVM classifier. To validate the effectiveness of the algorithm, we design a number of experiments. We illustrate the effectiveness of character representation, robustness of histogram equalization, influence of parameters and performance in severe situations by tables and graphs. Finally we analyze the experimental results in detail.In the last part we conclude the paper. All the work discussed in this paper is simulated under MATLAB 7.0 environment. The experimental results show that the ALBP method proposed in this thesis, compared to other LBP methods, show better performance in extracting the iris texture characteristics. The method is also robust against histogram equalization and different parameters cannot disturb the result to some extent. Furthermore, the algorithm also shows better performance in severe conditions than other methods using the same database. Therefore, the proposed method is tremendously competitive.