Fast Iris Localization Algorithm Onnoisy Images And 3D Reconstruction Based On CGA

Iris is a circular issue lying between pupil and sclera which has pigmented and uneven features. Scientists divided the iris into 30 regions where each region corresponds to an organ. Iridology analyzes shape, depth and location info of iris textures to diagnose diseases and determine well-being condition of a person. An iris diagnosis system utilizes computer technology to finish the process of Iridology. It can not only predict diseases, but also set up long-term database recording health conditions of patients. The difficulties include rapid localization of iris and feature extraction of texture position and depth info. In this paper we locate the iris under the conditions which there are a lot of noise in the iris image, then build the 3-dimensional model to obtain the depth information, next is to extract feature information using density-based clustering algorithmFirst, the raw iris images sampled by certain instrument are inevitably terminated by some noise, such as uneven lighting condition, light spots and occlusion by eyelid and eyelashes. Today, most of the algorithms separate the localization into 2 steps, coarse stage and fine stage, which leads o low localization speed, and low accuracy. To solve this problem, we proposed a method to make images three-valued. Through three parts partition-pupil area, iris area and sclera area, a major proportion of noise influence can be removed. Then, localize the iris inner and outer edge simultaneously by a CGA based circle detector. In CGA, a circle is formed by 3 points on it. So we choose 3 points randomly from the edge point set. A CGA based circle detection algorithm can detect several circle at the same time which leads to fast localiza1 tion speed and accurate localization.Second, since the conformal geometry algebra can represent geometric entities in a unified way, and the calculation between them is easy, so we proposed a CGA based 3D reconstructing algorithm. Use CGA sphere to represent the eyeball. We assume that the limbic boundary is fixed, so we just need to normalize the coordinate and transfer the coordinate into CGA space. And reflect it onto the eyeball model. Also, the depth information of the feature can be calculate through a depth and gray-value function. The model can show the iris in an intuitive way, and also helps to record the texture on the iris.Third, a density-based clustering algorithm can partition discrete data with natural relation but irregular distribution into one group. Integrate the advantage of such algorithms, we proposed a iris texture clustering algorithm to analyze and represent the feature on iris surface. First, we preprocess the image to eliminate the noise of light spots. Yet for the noise like eyelid and edge points of the iris, if not eliminated, will be clustered by mistake. So we solve this problem by adding constraint condition. Our purpose is to form different feature groups and analyze their position and area characters. Experiments shows that the algorithm can be used to record the features of each patient and compare the difference.