Face recognition based on an improved center symmetric local binary pattern
Face recognition based on an improved center symmetric local binary pattern
复制标题
基于改进的中心对称局部二值模式的人脸识别
DOI：
10.1007/s00521-017-2963-2
复制
发表时间：
2017
影响因子：
6
通讯作者：
Zhang Shaobai
中科院分区：
文献类型：
--
作者：
Zhou Ningning;Constantinides A.G.;Zhang Shaobai
提出了一种将中心像素信息融合到中心对称局部二值模式(CS-LBP)中的局部纹理特征描述子，用于人脸识别。CS-LBP由于其对光照变化的耐受性和计算效率，在人脸识别中得到了广泛的应用。但该算子完全忽略了中心像素信息，这在某些应用中可能会影响判别结果。为了利用更多的有用信息，本文将中心像素信息融合到CS-LBP描述子中，即CS-LBP/Center。在人脸识别中，首先将人脸图像分成小块，从中提取CS-LBP/Center直方图，然后用图像熵对其进行加权。最后，将所有的加权直方图串联起来，生成最终的人脸识别纹理描述符。在一些人脸数据集上的实验结果表明，采用最近邻分类方法可以获得较高的识别准确率。
This paper proposes a local texture feature descriptor which fuses the center pixel information into the Center-Symmetric Local Binary Pattern (CS-LBP) for the purpose of face recognition. Because of its tolerance to illumination changes, and computational efficiency, the CS-LBP is widely used in face recognition. But this operator completely ignores the center pixel information which may affect the discriminative result in some applications. In order to take advantage of more useful information, this paper fuses the center pixel information into CS-LBP descriptor, namely CS-LBP/Center. In face recognition, the face image is first divided into small blocks from which CS-LBP/Center histograms are extracted and then weighted by image entropy. Finally, all the weighted histograms are connected serially to create a final texture descriptor for face recognition. The experimental results on some face datasets show that a higher recognition accuracy can be obtained by employing the proposed method with nearest neighbor classification.