基于卷积神经网络的人耳识别研究  

Human Ear Recognition Based on Convolutional Neural Network

作　　者：胡颖[1,2] 穆志纯[2] 

机构地区：[1]山西职业技术学院电气工程与自动化系,山西太原030006 [2]北京科技大学自动化学院,北京100083

出　　处：《中北大学学报（自然科学版）》2015年第5期597-601,共5页Journal of North University of China(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61472031)

摘　　要：在分析人耳生物特征及其识别方法的基础上,提出了一种基于卷积神经网络的人耳识别方法.该方法直接提取人耳的图像特征,利用两次特征提取分类识别,在不同角度变化和不同维数情况下,嵌入维数越大,降维后所保留的特征信息越多,识别概率越大;维数的变化对于CNN识别没有任何影响.实验结果表明,该方法在人耳姿态变化时能够保持良好的识别概率,对人耳识别角度变化具有较好的鲁棒性.The human ear biometrics characteristics and recognition method was analyzed,and a new method was proposed based on convolution neural network（CNN）.The classification and recognition was made by using twice extractions of ear feature.In the conditions of different angles and different dimensions,the bigger embedding dimension was,the more retained feature information was after dimensionality reduction,and the greater the recognition probability was;the dimension change had no influence on CNN recognition.Experimental results show that this method can maintain good recognition probability at changing posture of human ear,and has good robustness at changing the recognition angle of ears.

关 键 词：人耳识别 卷积神经网络 感知神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]