LLE与核Fisher判别分析结合的人脸识别研究

Face Recognition Based on Locally Linear Embedding and Kernel Fisher Discriminate Analysis

针对人脸识别问题提出了将LLE与核Fisher相结合的识别方法 LLEKF,先应用LLE方法将样本和待测试的人脸图像集降低到一定维数,再利用核Fisher判别法通过选择合适的核函数,确定最优参数,对降维后的样本图像进行训练,并对降维后的人脸图像进行分类。实验证明,利用LLE低维嵌入后的数据能够更好地保持原人脸数据的非线性特征,并降低特征提取的时间,再经过核Fisher进行分类,明显提高了分类的效率。

A face recognition method that combines LLE and Kernel Fisher( LLEKF) was proposed.Firstly the sample and the testing face image were reduced to a certain dimension; secondly the sample image in reduced dimension was trained with selecting appropriate kernel function and determining optimal parameters by kernel Fisher.Finally the face image in reduced dimension was classified.Experiments show that LLE low dimensional embedding data is able to maintain nonlinear feature better in original face,and it reduces time in feature extraction.Kernel Fisher classification significantly improves the classification efficiency.