Ear Recognition Based On Modular Neural Networks

The Biometrics identification technology is a measure which using the physiology or behaviors characteristics of human being to identify personal. It provides a identify way with highly reliability and stabilization. The ear recognition as a newly branch in this field, at the same time, is one of the most challenging tasks in pattern recognition and computer vision. Comparing with other biometric technologies, the human ear has its own unique physiology and structure advantages, so that it getting more and more attention because of the potential application in public safety and information security fields. The measures are almost needed images preprocessing, complicate feature extraction, and select classifierin identification which use ear feature nowadays. The choice of using which features or classifiers will influence the final results of identification rate.The neural networks is a successful application in the pattern recognition.Comparing with traditional neural networks,the modular neural networks is strongly promoted in choosing parameters and learning samples added. This paper is focusing on the modular neural networks for human ear recognition, mainly study works are listed below:(1) All images in the human ear image library got preprocessing in this paper. The images are getting smoothing by median filtering and normalization by bilinear interpolation, to light effect by gray histogram equalization. The images are much more suitable for ear biometric recognition research.(2) Studying and researching the theory of principal component analysis, and use it to the human ear image feature extraction.(3) Studying and researching the neural networks to explain the concept of “modular”, this paper proposed the structure design and advantages of modular neural networks. This paper discussed the feasibility of human ear recognition usingmodular neural networks, and analyzed the improvement direction and reasons of error identification.(4)In order to improve the accuracy of identification, this paper proposed the optimization of modular neural networks using hierarchical genetic algorithms(HGA).By optimizing the structure and design parameters of MNN to improve the recognition rate for the human ear.The trained network to the human ear recognition experiment, the combination of two methods not only improves the accuracy but also prove the feasibility.