小样本条件下基于全局和局部特征融合的人脸识别

Global and Local Feature Extraction Based Face Recognition in Small Samples

针对线性判别分析的小样本空间问题,提出了一种基于类向量的融合全局和局部特征的人脸识别算法。首先,提取人脸的全局特征;然后将人脸分割成6个关键部分,并用一种新的基于Gabor小波的方法提取特征;其次,将全局和局部特征融合,得出样本的特征向量;再次,得出每类样本的类向量并据此得出一种新的投影准则;最后,将类向量和试验样本分别进行投影,根据其欧氏距离的大小得出试验人脸的最终类。试验表明本文算法不仅能有效解决小样本空间问题,而且计算速度快,识别率高,应用前景良好。

A new algorithm of face recognition based on global and local feature extraction was proposed to solve small samples in LDA. Firstly,the global feature was extracted and the feature of six key parts test face divided was extracted through Gabor wavelet. Then, global and local features were fused to get eigenvector of samples. Class vector of each kind of samples was calculated, and accord- ing to which, a new project rule was gotten. In the end,class vector and test samples were projected separately. The final class test face belonged to was declared by Euclidean distance. The experiments show that the proposed algorithm can deal with small samples problem effectively, and is fast, high recognition rate.