基于局部特征约束的多样本判别字典学习算法

Sample Diversity Discriminative Dictionary Learning Algorithm Based on the Locality Constrained

为了提高人脸识别的性能,提出基于局部特征约束的多样本判别字典学习算法。首先,利用人脸的对称性特征设计虚拟样本,并结合原始训练样本构造多样本判别字典学习算法。其次,利用原子间的相似性特征,构造基于原子局部特征约束的判别式模型,促使字典继承训练样本的局部特征,增强字典的判别性能。在LFW和FERET两个人脸数据库中,实验结果表明所提出的算法比5个稀疏编码和字典学习算法取得更高的分类性能。

In order to improve the performance of face recognition, proposes a sample diversity discriminative dictionary learning algorithm based on the locality constrained. According to the face geometrical property, it first produces virtual face images and then designs a sample diversity discriminative dictionary learning algorithm. Then, constructs a discriminative term based on the locality constrained by using the atoms to inherit the locality characteristic of training sample. Thus, it can enhance the discriminative ability of the learned dictionary. Experiment re- suits demonstrate that the proposed algorithm achieves better classification results than five dictionary learning and sparse coding algo- rithms on the LFW and FERET face databases.