基于鲁棒的局部二值模式人脸识别算法 A Novel Face Recognition Algorithm Based on Robust Local Binary Pattern

本文针对LBP算法特征包含outlier和维度过高的问题提出了一种基于鲁棒的局部二值模式(RobustLBP)的快速有效的人脸识别算法。RobustLBP算法的思想是在LBP算法的基础上加上一个Robust函数除去outlier达到降维的目的。首先通过计算LBP特征各个维度和中心元素的马氏距离作为Robust函数的输入，使得Robust函数收敛估算出重要信息。然后利用这些信息求出变换矩阵除去原始LBP特征的outlier。最后比对降维后特征间的卡方距离实现人脸识别。在FERET、CAS-PEAL-R1、LFW人脸数据库上的实验证明本文提出方法在是人脸识别上具有优越性。 This paper is aimed at solving the problems that LBP feature contains outlier and the dimension of LBP fea- ture is too high, and a fast and effective face recognition algorithm based on Robust Local Binary Pattern is proposed. The main idea of RobustLBP is setting a Robust function on the basis of original LBP. First, it calculates the Maha- lanobis distance between the mean vector and every dimension as the argument of Robust function and estimates a set of important information by making Robust function convergence. Then, it obtains a transformation matrix which is used to reject outlier of original feature by using the information. Lastly, it compares the Chi-square distance among the features after reducing dimension in order to complete face recognition. Extensive experiments on FERET, CAS- PEAL-R1 and LFW face databases validate the effectiveness of face recognition.