基于2DNPP和Trace变换的平面内旋转人脸识别  

In-Plane Rotary Face Recognition Based on 2DNPP and Trace Transform

作　　者：施展[1] 杜明辉[1] 梁亚玲[1] 

机构地区：[1]华南理工大学电子与信息学院,广东广州510640

出　　处：《华南理工大学学报（自然科学版）》2012年第8期46-50,共5页Journal of South China University of Technology(Natural Science Edition)

基　　金：NSFC-广东省自然科学联合基金资助项目(U0735004);广东省自然科学基金资助项目(05006593);华南理工大学中央高校基本科研业务费专项资金资助项目(X2dXD2116370)

摘　　要：针对平面内具有随机旋转角度的人脸图像难以识别问题,提出一种融合二维近邻保持投影(2DNPP)和Trace变换的方法,以实现图像旋转不变性特征提取和识别.首先对图像做一重和二重Trace变换,然后对二重曲线进行匹配计算,得到既对平面内旋转变化具有鲁棒性、又能保存丰富图像信息的特征,最后通过2DNPP进行降维并分类.用该方法分别对正面的、旋转的、加噪声的人脸图像进行了识别实验,并与SIFT、pseudo-Zernike等方法进行了比较,结果表明:对于具有随机旋转角度的ORL图像库,文中算法识别率达到96%,且对白噪声具有较强的鲁棒性.As it is difficult to recognize the facial image with a random rotation angle in a plane, a method integra- ting 2DNPP with Trace transform is proposed for the extraction and recognition of rotation-invariant features. In this method, first, the image is dealt with through the first and the second Trace transform. Then, the matching calcula- tion of the double Trace feature is carried out, thus achieving a feature, which not only is robust to in-plane rotation but also can preserve most information of the raw image. Finally, the classification and dimension reduction are conducted by means of 2DNPP. The proposed method is compared with SIFT, pseudo-Zernike methods on frontal, rotary and noisy facial image databases. The results show that, on the ORL image database with a random rotation angle, the proposed method has a recognition rate of 96% as well as a strong robustness to white noise.

关 键 词：人脸识别 Trace变换 旋转不变性 特征提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]