图像多模态扰动的人脸识别方法  

Muiti-modal disturbing face recognition algorithm

作　　者：张国庆[1,2] 王正群[2] 王颖静[2] 徐伟[2] 

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094 [2]扬州大学,信息工程学院,江苏扬州225127

出　　处：《计算机工程与应用》2013年第7期204-207,247,共5页Computer Engineering and Applications

基　　金：国家自然科学基金(No.60875004);江苏省自然科学基金(No.BK2009184);江苏省高校自然科学基金(No.10KJB510027)

摘　　要：为了克服因人脸图像检测引起的配准不稳定性和小样本引起的维数灾难,由一副二维人脸图像通过上下左右平移生成4个图像,把生成的图像与原来的图像一起加入训练样本集,构成新的训练图像集。基于二维图像,结合图像局部结构信息,设计了准则函数,获得双投影矩阵,抽取人脸特征。对待识别人脸图像,由它的扰动图像设计识别方法。与传统的人脸识别方法相比,该方法的识别效果更好;Yale和ORL人脸数据库上的实验结果验证了该方法的有效性。In order to overcome the instability caused by face detection and curses of dimensionality resulted from small sample size, four new images from one two-dimensional image is generated by horizontal and vertical translation, and a new image data- base is formed by adding these new images to original image database. Combined with local structure information, this paper designs objective function and obtains dual projection matrix based on two-dimensional images. The proposed method is carried out on disturbing image database. Compared with the traditional face recognition method, it has a better recognition perfor- mance, and the experimental results on Yale and ORL face image database show that it is effective and robust.

关 键 词：二维线性判别分析(2DLDA) 小样本问题 图像扰动 特征抽取 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]