Research Of Iris Recognition Based On Empirical Mode Decomposition

With the development of the informational, digital and networked society, the demands on the security in the world and social life are put forward. For that reasons, the traditional security technique shows the unresolved and severe limitation. However, the biometric recognition technique is one of the best methods to improve the security of the informational, digital and network society. Since such features as the uniqueness, stability, high reliability and non-invasion, the iris recognition attracts the more and more attention. The representation and extraction of the iris's texture feature are the key and difficult point. Empirical mode decomposition (EMD), firstly introduced by Huang et al. in 1998, can adaptively decompose a signal into several intrinsic mode functions (IMF) with the frequency form the high to the low. As a data driver method, which analyzes the signal from its own scale feature, is locally adaptive. It is widely applied in the signal de-noising and fault diagnoses recent years. Nunes et al. and other scholars have extended this method to bi-dimension, and applied it in the texture analysis. This paper discusses how to apply EMD to the iris texture analysis with the experiment as the illustration.First of all, the paper briefly introduces EMD decomposition algorithm and bi-dimensional EMD framework and some key issues. Then it discusses the preprocess of the iris, focuses on the iris location, de-noising, normalization and enhancement, and improves Canny operator and the Hough Transform so as to reduce the memory consumption and save the time. This paper centered on the analysis and the feature extraction using EMD: (1) Adopt one-dimensional empirical mode to decompose the iris texture, regard the first intrinsic mode function as the noise, make the feature analysis on the first residual, and make the match experiment by the means of various distance measures. The results from the experiment show that experience mode decomposition can be effectively extracted and expressed the iris texture feature. (2) From the multi-scale characteristics of decomposition, the iris recognition method based on IMF is proposed in order to reduce the length of the encoding. (3) The empirical mode decomposition is extended from one-dimensional to bi-dimensional. Based on bi-dimensional empirical mode decomposition and local binary patterns, an iris recognition algorithm is proposed. It extracts local binary feature from the fist residual, and uses the local binary pattern histogram matching for classification and identification. Experimental results show that the method is robust, precise, superior in the speed and stable in the rotation.Most of the experiments are carried out on software platform of MATLAB 7.0 and experiment results are analyzed and summarized at last. In the Visual Studio development environment, some of the relevant algorithm was implemented and compared. The experimental results show that EMD used in the iris texture analysis has a good performance. And the algorithm proposed can be effectively applied to practical demands.