基于多种LBP特征集成学习的人脸识别*

单一的特征与分类器只能对限定条件下的人脸进行较好的识别,当在非限定条件下(如光照、背景等发生变化时)将出现人脸识别率较低问题,针对该问题,提出了一种基于多种局部二进制特征集成学习的人脸识别算法。首先,使用监督梯度下降法 (SDM)对人脸特征点定位,应用中心对称局部二进制(CSLBP)算子提取每个特征点邻域特征,将所有人脸特征点邻域特征合成为精细的纹理特征；同时运用分区LBP直方图算法提取人脸区域的微观空间结构特征；然后,使用K最近邻算法(KNN)和支持向量机(SVM)分别训练这两种特征,得到类别排序列表和投票决策矩阵；最后,利用加权求和的规则融合决策矩阵,构成最优集成分类器,从而得到输出类别。通过在非限制性人脸库LFW上实验结果表明,所提算法采用集成的方法明显优于单一的特征和分类器。

Face recognition based on ensemble learning with multiple LBP features

Single feature and classifier can better recognize the face under limiting conditions. When non-limiting conditions (such as light, background and pose etc.) change, it will appear lower recognition rate. So in view of the above problem, a new face recognition method based on ensemble learning with multiple LBP features was proposed. First, after locating the facial feature points by the Supervised Descent Method algorithm (SDM), The neighborhood features of each feature points are extracted by using the center symmetric local binary pattern (CSLBP) . The feature of all facial feature points is synthesized into fine texture features; The LBP histogram algorithm is used to extract the micro spatial structure features of human face region. Then the K nearest neighbor algorithm (KNN) and the support vector machine (SVM) is used to train these two kinds of features, which get the classification list and voting decision; Finally, an optimal ensemble classifier is constructed by using the weighted sum rule fusion decision matrix to obtain the output class. Experimental results based on non-limiting face database LFW show that the proposed algorithm is significantly better than the single feature and classifier.