摘要
文中将一种新颖的核学习算法—核最近邻凸包分类算法用于人脸的小波特征识别。该算法的设计受到支持向量机几何解释启发,利用核函数方法将数据映射到高维核空间,并在核空间构造以训练集凸包为扩展类集的最近邻分类器。文中采用的人脸图像的小波低频特征对人脸识别十分有效。人脸的小波低频特征不但保留了人脸的主要信息,而且具有较少的维度。在ORL人脸图像库上的“leave-one-out”测试方法的实验中,这种基于小波低频特征的核最近邻凸包分类算法取得了99.25%的识别率。
A new kernel learning method called Kernel Nearest Neighbor Convex hull （KNNCH） algorithm is used for wavelet features based face recognition. Inspired by the intuitive geometric interpretation of SVM based on convex hulls, KNNCH maps the data in the original space to the kernel space with the kernel trick and constructs a nearest neighbor classifier in the kernel space, which takes the convex hulls of training sets as the extended class sets. The lower frequency features of face images extracted by 2D wavelet transform are efficient for face recognition. The features not only preserve the main information of face images, but also have the less dirnensionality. KNNCH with wavelet features for face recognition shows very good performance, which can achieve 99. 25 % recognition rate with ＂leave- one -out＂ test method on ORL face database.
出处
《计算机科学》
CSCD
北大核心
2007年第5期224-227,共4页
Computer Science
基金
国家自然基金资助项目(No.60472060)
关键词
核学习
核最近邻凸包
小波变换
模式识别
人脸识别
Kernel learning, Kernel nearest neighbor convex hull, Wavelet transform, Pattern recognition, Face recognition
