基于DCT与LDA的仿生人脸识别研究

摘要

A new method of biomimetic pattern face recognition theory based on DCT and LDA is proposed. This method has solved the low recognition rate and the excessively high dimension problem. The features of human face on the training samples are extracted through DCT and LDA, which are mapped into the high-dimensional space through kernel function and the high-dimensional space characteristic is used to construct the cover region of each kind of samples. The human face is distinguished through the judgment that the human face characteristic belongs to which kind of cover region or doesn't belong to any region. The experiment on Yale and ORL face library demonstrates this method achieves much better results in the efficiency and the feasibility of human face recognition.%针对基于DCT变换与LDA的人脸识别方法识别率低和特征提取过程中维数也低,以及基于K-L变换的仿生人脸识别方法识别率高和特征提取过程中维数也过高的问题,结合两者的优点,提出了一种基于DCT与LDA变换的仿生人脸识别的方法.通过DCT变换与LDA对训练人脸样本进行特征提取,通过核函数将提取的特征映射到高维空间,构建各类样本的覆盖区域,再通过判断待识别人脸特征在各覆盖区域的归属情况来识别人脸.在Yale和ORL人脸库上的实验证明提出的方法取得了较好的识别效果.