A method that combines the nonlinear down-dimentioned method with Fisher linear discrimination is presented to improve the recognition rates of face recognition algorithm based on popular learning theory. Firstly, the dimensions of face image tested and training set data are reduced to an appropriate dimensionality through threshold embedding algorithm, and then the Fisher linear discrimination is used to extract the face features. Finaly, the features of testing and training face images are classified by nearest neighbor classifier. The comparison between the face recognition algorithm based on LLE and Fisher linear discrimination and the typical face recognition algorithm based on the popular learning theory was conducted in Olivettifaces and ORL face image databases. The results show that the algorithm proposed in this paper has the highest recognition rate compared with some classical methods when the number of the nearest neighbor is large.%为了提高基于流形学习理论人脸识别算法的识别率,采用一种将非线性降维与Fisher线性判别相结合的方法.首先利用邻域嵌入算法,将人脸图像测试和训练集的维数降低到合适维度,然后使用Fisher线性判别进行人脸数据集特征的提取,最后将测试集人脸图像特征和训练集人脸图像特征,使用最近邻分类器进行分类.在公开的Olivettifaees和ORL人脸图像数据库上,分别将该算法与几种经典基于流形学习理论的人脸识别算法进行了对比实验,实验结果表明当近邻数比较大时本算法识别率是最高的.
展开▼
