基于LLE的多姿态人耳识别  

Multi-pose ear recognition based on locally linear embedding

作　　者：谢朝霞[1] 穆志纯[1] 谢建军[2] 

机构地区：[1]北京科技大学信息工程学院,北京100083 [2]河南科技大学机电工程学院,河南洛阳471003

出　　处：《智能系统学报》2008年第4期321-327,共7页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(60375002,60573058);北京市教委重点学科共建项目(XK100080537)

摘　　要：多姿态人耳识别是人耳识别技术面临的一个难题,目前这方面的研究并不多见.通过分析国内外2-D人耳识别方法在解决姿态问题时存在的不足,引入流形学习算法,提出一种基于局部线性嵌入的多姿态人耳识别方法.实验结果表明,这种方法在人耳姿态变化时能够取得非常理想的识别率,提高了人耳识别的鲁棒性,增强了人耳识别技术的实用性.Multi-pose ear recognition is a challenging problem in ear recognition technology, and it has not received sufficient attention. In this paper, on the basis of the manifold learning algorithm, we propose a multi-pose ear recognition method based on LLE （locally linear embedding ）that overcomes the disadvantages of 2-D ear recognition methods in dealing with pose variations. Experimental results show that this method can obtain a satisfactorily high recognition rate, improving the robustness of ear recognition, and enhancing the practicability of ear recognition technology.

关 键 词：多姿态人耳识别 流形学习 局部线性嵌入 姿态变化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]