Multiple Features Of Ear Recognition

Ear recognition is an active research area in pattern recognition. It has very important significance to the pattern recognition development and the application. We make research in the three steps of Ear recognition question, such as: Ear feature extraction; Eat feature fusion and feature selection; Ear recognition at last. Our research work focus on the following aspect:(1)We adopt a Ear feature fast extraction on fast zernike wavelet moments algorithm.The many kinds of Ear features extraction are together extremely consume the time,in order to enhance the action rate of the Ear recognition System, We compare the zernike moment algorithm and the wavelet moment algorithm, by adopting the two algorithms, simplified the algorithm step, we enhance the speed of zernike wavelet moments algorithm at last.(2) A method of feature extraction based on PoSVD is proposed in the paper.In order to extract a kind of feature that can reflect the Ear picture texture distribution characteristic better, a adopted PoSVD method which is extracting SVD feature from polar transformation matrix of picture is proposed. This method alse has the very good compatibility to the picture shooting angle factor influence.(3)In the step of Ear feature fusion and feature selection,We has used the feature level serial fusion method to fuse the zernike wavelet moment feature and PoSVD feature completely. So that the valid information of the features is kept and the redundant information is eliminated.In order to reduce the dimension of the fusion feature, a method of feature selection based on PIDF Dynamic plan which is called PIDFDP in this paper is proposed.This method made the improvement after the PIDC method to use in the feature selection, has very strong surveillance ability for feature selection.(4)In the step of Ear recognition, a classification based on Biomimetic Pattern Recognition which can classify target by High Dimensional Space Geometry Covering is used. This method overcome that the traditional matter classification is high in error.In order to improve the precision of the Neural network based on High Dimensional Space Geometry Covering, a Unsupervised combination two-weight Neural network classifier is proposed. This method use K-means clustering to block the breakdown of the sample, then use the block sample to guide to neural-network classifier training and combinating. It improve the recognition rate of classification.Using the methods of Ear feature extraction,Eat feature fusion and feature selection,Ear recognition in this paper, it can achieve rapid and accurate Ear recognition based on passive Ear Images ,and can be a very good result.