自适应阈值局部特征融合的人脸识别

Face recognition based on adaptive threshold local feature fusion

针对局部二值模式(local binary pattern,LBP)提取图像纹理特征时阈值不能自适应以及缺少方向信息的问题,文章提出了一种自适应阈值局部特征融合的人脸识别算法。首先对原始人脸图像进行分块处理,提取每个子块自适应阈值均匀模式的局部二值模式(uniform local binary pattern,ULBP)特征和局部梯度编码(local gradient coding,LGC)特征;然后,将2种局部特征串联在一起融合为1种特征,得到每个子块的直方图特征,把每个子块图像的信息熵作为直方图加权系数,将所有子块图像的直方图乘以各自的加权系数,连接得到整幅人脸图像的直方图特征;最后用支持向量机(support vector machine,SVM)分类器进行识别分类。在ORL、Yale、FERET(be、bj、bf子库)人脸数据库上进行试验,该人脸识别算法分别得到了99.0%、98.7%、87.5%、93.0%、88.5%的识别率,正确识别率较高,算法对其他纹理分类、目标识别也具有一定的参考价值。

In order to solve the problems of the threshold value not being adapted and lacking of directional information in extracting image texture features from local binary pattern(LBP),a face recognition algorithm based on adaptive threshold local feature fusion is proposed.Firstly,the original face image is divided into several blocks,in which uniform local binary pattern(ULBP)with adaptive threshold and local gradient coding(LGC)operator are used to extract features.Then two features are fused together in series as a feature in order to obtain histogram feature of each sub block.The information entropy of each sub block is calculated as a histogram weight,the histogram feature of whole image is generated by connecting information entropy with histogram feature of every sub block.Finally,faces are classified by support vector machine(SVM)classifier.On the ORL,Yale,FERET(be,bj,bf)face databases,the recognition rate of the proposed algorithm is 99.0%,98.7%,87.5%,93.0%and 88.5%,respectively.The proposed algorithm can obtain higher correct recognition rate and it has reference value for other texture classification and target recognition.