一种基于联想记忆神经网络的单训练样本人脸识别算法  

An Algorithm for Face Recognition with Single Training Image per Person Based on Associative Memory Neural Network

作　　者：陈蕾[1] 龙波[2] 王传栋[1] 

机构地区：[1]南京邮电大学计算机学院,江苏南京210003 [2]江西行政学院现代科技教研部,江西南昌330033

出　　处：《微电子学与计算机》2009年第10期213-216,共4页Microelectronics & Computer

基　　金：国家自然科学基金项目(60703086)

摘　　要：通过在多值自联想记忆神经网络(MVAM)中引入大脑皮层中客观存在的"侧反馈"连接机制,提出了一种具有联机学习能力的基于侧反馈机制的多值自联想记忆模型(MVAM-LF),用于解决单训练样本人脸识别问题.与一般的人脸识别方法不同,本模型无需任何人工的几何特征提取工作.在部分FERET人脸数据库上的随机加噪和部分遮挡的识别实验表明,提出的MVAM-LF算法获得了比标准的eigenface和改进的(PC)2A算法更高的识别率,表现出了较强的鲁棒性.The multiple-value auto-associative memory neural network with lateral feedback connection （MVAM- LF） is proposed for face recognition with single training image per person. MVAM-LF is a modified multiple- value auto-associative memory model （MVAM） with online learning ability by introducing ＂lateral feedback＂ connection mechanism which objectively exists in the cerebral cortex, Distinguished from the conventional face recognition methods, this model do not need any manual geometric feature extraction. Experimental results on part of the FERET face image database show that compared with standard eigenface （PCA） and modified （pC）^2A algorithm, the proposed algorithm has relatively high recognition ratio after adding salt ＆ pepper noises and partly occluded in a mosaic way on face image.

关 键 词：人工神经网络 联想记忆 侧反馈机制 单人单训练样本 人脸识别 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]