Abstract: In order to improve the recognition rate of ear, a novel ear recognition algorithm is proposed based on improved locally linear embedding algorithm. Firstly, the ear features are extracted by Curvelet transform, and then the improved LLE is used to reduce dimensionality of features and select the optimal features of ear recognition, finally RBF neural network support vector machine is used to establish the classifier of human ear recognition, and the simulation experiment is carried out on USTB3 ear images. Compared with the reference methods, the proposed algorithm has obtained higher ear recognition rate, and the experimental results verify the effectiveness.

Key words: multi-ear, curvelet transform, features extraction, locally linear embedding, RBF neural network

摘要： 为了提高多姿态人耳识别的准确性，提出了一种改进局部线性嵌入算法的多姿态人耳识别方法。通过曲波变换提取人耳特征，采用改进局部线性嵌入算法对特征进行降维，选择最有利于人耳识别的特征向量;采用RBF神经网络建立人耳分类器实现人耳识别，并采用USTB人耳图像库进行仿真实验。相对于其他方法，该方法获得了更高的人耳识别率，从而验证了算法的有效性。

关键词: 多姿态人耳, 曲波变换, 特征提取, 局部线性嵌入, RBF神经网络