稀疏表示及其在人脸识别中的应用

稀疏表示是近年来新兴的一种数据表示方法,是对人类大脑皮层编码机制的模拟。稀疏表示以其良好的鲁棒性、抗干扰能力、可解释性和判别性等优势,广泛应用于模式识别领域。基于稀疏表示的分类器在人脸识别领域取得了令人惊喜的成就,它将训练样本看成字典,寻求测试样本在字典下的最稀疏的表示,即用尽可能少的训练样本的线性组合来重构测试样本。但是经典的基于稀疏表示的分类器没有考虑训练样本的类别信息,以致被选中的训练样本来自许多类,不利于分类,因此基于组稀疏的分类器被提出。组稀疏方法考虑了训练样本的类别相似性,其目的是用尽可能少类别的训练样本来表示测试样本,然而这类方法的缺点是同类的训练样本或者同时被选中或者同时被丢弃。在实际中,人脸受到光照、表情、姿势甚至遮挡等因素的影响,样本之间关系比较复杂,因此最后介绍局部加权组结构稀疏表示方法。该方法尽量用来自于与测试样本相似的类的训练样本和来自测试样本邻域的训练样本来表示测试样本,以减轻不相关类的干扰,并使得表示更稀疏和更具判别性。

Reviews of Sparse Representation and Its Applications in Face Recognition

Sparse representation is one of the hottest data representation methods which is thought to underlie the neural representations used by brain, and it has been developed sound theoretical foundation. In recent years, sparse representation based classification (SRC) has led to the interesting face recognition results and it looks for the sparsest representation of a query face image with respect to a dictionary composed of all the training images. However, the -norm regularized sparse representation is not stable and fails to incorporate the label information of training samples. Group sparse classification (GSC) extends SRC, according to label information, the training samples are grouped. GSC only selects a few groups to represent the query sample by using an -norm regularization. However, for a particular group, all the training samples are selected. A new classification method called weighted group sparse representation classification(WGSRC) to classify a query image by minimizing the weighted mixed-norm (-norm) regularized reconstruction error with respect to training images is proposed. WGSRC gives each group a weight. We try to represent a test sample by training samples not only from the neighbors of it, but also from the highly relevant classes.