曲率与小波轮廓增强的人脸识别算法  

Face recognition algorithm based on curvature and wavelet contour enhanced

作　　者：周先春[1,2] 唐娟[1,2] 汪美玲[1,2] 孙文荣[1,2] 

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044 [2]南京信息工程大学江苏省大气环境与装备技术协同创新中心,江苏南京210044

出　　处：《电子技术应用》2015年第10期161-164,共4页Application of Electronic Technique

基　　金：国家自然科学基金项目(11202106);教育部高等学校博士学科点专项科研基金项目(20123228120005);江苏省高校自然科学研究项目(13KJB170016)

摘　　要：为了克服非约束性变化条件下人脸识别率降低的弊端,提出一种曲率与小波轮廓增强的人脸识别算法。首先建立结构控制函数,通过水平集曲率检测人脸图像的整体结构,并建立融合轮廓分布模型,得到融合分布图像。然后用小波增强融合分布图像,得到轮廓和整体结构增强的图像,在此基础上,用主成分分析(PCA)算法对上述增强图像进行特征提取。最后通过稀疏表示(SRC)判断测试图像所属的类。实验结果表明,在ORL数据库的基础上,与PCA识别算法、SRC识别算法以及PCA与SRC相结合(PCA&SRC)的识别算法相比,该算法在非约束条件下识别率最高,鲁棒性得到增强。In order to overcome the drawback that recognition rate declines sharply under the condition of non-constraint, a face recognition algorithm based on curvature and wavelet which is used for contour enhancement is proposed. Firstly, a structure control function is established, which uses the level set curvature to detect the overall structure of the face images, and a fused contour distribution model can be built to get a fused distribution image. Then, wavelet is used to enhance the fused distribution image, and obtain the image with enhanced contour and overall structure, the principal component analysis （PCA） algorithm is used to extract the feature of the enhanced image. Finally, the sparse representation is used for judging the classification of the testing image. Based on the ORL database, the experimental results indicate that the proposed algorithm has a better recognition rate andS＇robust performance than other mentioned algorithms, such as PCA algorithm, SRC algorithm and PCA ＆ SRC algorithm which is the combination of PCA and SRC.

关 键 词：人脸识别 稀疏表示 主成分分析 水平集曲率 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]