一种相似性学习算法及其在人脸识别中的应用

传统的支持向量机相似性学习算法在构造样本对时,会考虑所有的原始训练样本,致使样本对空间和原样本空间呈平方关系,而过多的训练样本对会降低训练速度。为此,提出一种改进的支持向量机相似性学习算法,并应用到人脸识别中。引入二元样本对方法构造样本对,采用K近邻算法减少不相似样本对的生成,从而加快支持向量机的训练速度,同时使用随机降维方法来降低人脸数据的维数。实验结果表明,与基于差空间样本对和差绝对值样本对的算法相比,该算法可获得更高的识别率。

A Similarity Learning Algorithm and Its Application to Face Recognition

Traditional similarity learning algorithms consider all training samples when constructing paired-samples, which would lead to a larger paired-sample space that depends on the training samples in a square fashion. It is well-known that it is time-consuming when Support Vector Machine（SVM） processes a large-scale classification problem. Aiming at this problem, this paper proposes an improved similarity learning algorithm using SVM, and applies it to face recognition. This paper introduces a new paired-sample construction method, called pairwise sample method. In order to speed up the training procedure, it adopts K Nearest Neighbor（KNN） algorithm to reduce the number of dissimilar paired-samples. In addition, the random projection method is introduced to reduce the dimensionality of face data. Experimental results show that the improved algorithm has better classification performance compared with algorithm based ondifference space paired-sample and algorithm difference absolute value paired-sample.