Research On Key Technologies Of Iris Image Preprocess And Feature Extraction

Through long term research of human iris, both Traditional Chinese Medicine's and Western Medicine's Iridology think that iris tightly correlate with other apparatus. The changes of iris directly reflect health condition of apparatus and cure process. The purpose of this paper is labeling regions whose texture structure is changed from iris image and finding out some texture definitions and templetes to classify normal and disease regions. In this paper, we mainly take Autonomic Nerve Wreath, Radii Solaris and normal region as research subject. The main research contents are as follows:At first, we introduce preprocess of iris image in detail. This paper presents adaptive gamma correction method based on characteristic pattern. This method takes characteristic pattern of iris image as basis of gamma correction and considers distance. We calculate gamma value using simulated annealing algorithm. In this paper, pupil partition method is used for rough location and transfiguration template is used for iris precious location. In the end, circularity iris is normalized to rectangle with fixed size by mapping Cartesian coordinates into polar coordinates.Secondly, we define pattern class of Autonomic Nerve Wreath, Radii Solaris and normal region definitely. And then we describe a sample extraction method based on sliding window. This method partitions iris image after preprocess and then chooses samples from samples based on above different iris region definition. In this paper, statistic methods such as GLCM, fractal and Hu invariant moments are used for describing samples'texture feature. In allusion to texture of samples, we extract texture features at both original and wavelet image.And then, we present a method for feature selection based on Self-Organizing Feature Map (SOM). The method makes cluster result of SOM Network best by selecting a feature set using greedy arithmetic. The result of SOM Network shows different class samples are clustered to different output level nerve cells. The result shows the three classes have good distinctiveness.We analyse and evaluate selecting features using relative entropy, and explain the result of feature selection reasonably.Finally, we classify unlabeled samples of nervous system diseases, digestive system disease and health using SVM and SOM Network at the same time. The result shows that we can classify Autonomic Nerve Wreath, Radii Solaris and normal region accurately, and lay the foundations of primary diagnosis.