Algorithm Research On Ear Characteristic

Ear Recognition is an important biometric identification method. It has a wide range of uses. It is feasible to distinct different person by person's ear. Ear recognition has an unparalleled advantage than the other methods. The most important thing of ear recognition is extracting simple and effective feature, which is used to distinguish different person. Unified theory has not been raised. We discuss the current methods on how to get human ear features. And then we propose a new category method of forming ear features: feature forming methods based on non-planar curves and contours in plane. We will discuss in the following three areas on this topic.First, this paper puts forward six new methods on how to process detecting image of ears. Their core idea is extracting the similar part from different ears. Ear recognition could become meaningful only in this way. It should be stated that the first four algorithms is followed, and the last two algorithms is parallel. We should process the ear image by the first four algorithms, and then select the fourth or the fifth algorithm according to the actual needs. These methods are as follows: extracting the long axis of ear curve and marking endpoint, determining the two vertices of ear edge, determining the two nodal which is created by the long axis of ear curve and ear contours, eliminating the black pixels of the four ear vertexes, detecting the internal and external ear contour line, dividing the edge sequence by equal interval, and so on.Secondly, we give the definition of the method which could format ear characteristics by non-planar curves and contour: extract the outer contour line and internal counter line, and then differentiate the extracting curve by using a certain type of recognition method based on non-planar curves and contour, thus we could complete the task of recognition of the original ear image. We discuss these methods in deep after we give the definition of feature forming methods based on non-planar curves and contours in plane. There are many types of such methods. We give several new concrete algorithms under the guidance of this method in this paper. These methods are ear characteristics formatting method based on curvature of ears, ear characteristics formatting method based on line ratio, ear characteristics formatting method based on Hu contour moment invariants, ear characteristics formatting method based on the angle between the long axis and centre, ear characteristics formatting method based on spindle rate. What we need to emprise is that these methods transform the ear recognition thoroughly.Finally, some classic methods will be applied to the identification of the human ear in order to enhance the effectiveness of features. Three feature compressing methods are given. These methods are ear feature extracting algorithm based on Euclidean distance, the searching algorithm of ear feature based on limited branch, ear feature selecting method based on increasing l characteristics, reducing r characteristics. The description of above algorithms includes three parts mostly. The first part is algorithm rationale. We prove the invariability of revolving, translating, and expansing. In the second part, we give the concrete realization, we give the detailed step. In the third part, we give the explanation to certain steps of the algorithm and point out the reason of using these steps. In the last part, we give the procedure code of the first algorithm and the result of one person's ear by using the first algorithm.The extracting and compressing methods of ear features which are proposed in this paper have a certain theoretical and practical significance.