基于分块加权的局部保持投影的人脸识别

Locality Preserving Projection for Face Recognition Based on Weighted Sub-patterns

为增强对姿势、表情、光照等变化的鲁棒性,提出了一种加权的分块局部保持投影人脸识别算法.算法先对样本图像分块,对分块得到的子图像利用局部保持投影算法分别提取局部特征信息,并利用k近邻点的类标信息和样本影响力函数计算各分块子图像的权重.该算法能够有效地抽取图像的局部特征,对人脸表情和光照条件变化较大的图像表现尤为突出,在AT&T和Yale人脸库上的比较实验说明了该算法的有效性.

To enhance the robustness of facial pose, expression and illumination variation, a weighted sub-pattern locality preserving projection algorithm is proposed for face recognition. First, the original whole face images are divided into non-overlapped modular images, which are also called sub-patterns. Then, the locality preserving projection algorithm is directly used to the sub-images obtained from the previous step. And the contribution of each sub-pattern can be computed through the label information of k-neighbor and sample influence function. In face recognition, it is really true of the images that have large variations in facial expression and lighting, by this way, the local feature of the images can be extracted efficiently. To test weighed modular LPP and to evaluate its performance, a series of experiments are performed on two human image databases： Yale and AT＆amp;T human databases. The experimental results indicate the effectiveness of weighed modular LPP.