文档介绍：第 27 卷第 4 期计算机辅助设计与图形学学报 Vol. 27
2015 年 4 月 Journal puter-Aided Design & Computer Graphics Apr. 2015
LBP 和 HOG 的分层特征融合的人脸识别
万源1), 李欢欢1)*, 吴克风2), 童恒庆1)
1) (武汉理工大学理学院武汉 430070)
2) (华中科技大学自动化学院武汉 430074)
(whlg1204lhh@)
摘要: 针对 LBP 描述子提取的纹理特征有限且不能有效地描述图像边缘和方向信息的问题, 提出了 LBP 和 HOG
的分层特征融合的方法. 首先利用 LBP 算子提取图像的分层纹理谱特征, 然后利用 HOG 算子提取原始图像的边缘特
征和基于分层 LBP 特征的分层 HOG 特征, 最后将分层 LBP 特征分别与 2 种 HOG 边缘特征融合, 得到 2 种不同的多
层融合特征. 通过在 ORL, Yale 和 GT 人脸库上进行实验, 比较了 15 种算法的识别性能, 结果证明了文中方法的有效
性; 相对于传统的经典降维算法、单一的 LBP 特征提取算法和 HOG 特征提取算法, 该方法的识别率有很大的提高,
分别达到 99%, %和 %.
关键词:人脸识别; 局部二值模式; 梯度方向直方图; 分层特征; 特征提取
中图法分类号:
Fusion with Layered Features of LBP and HOG for Face Recognition
Wan Yuan1), Li Huanhuan1)*, Wu Kefeng2), and Tong Hengqing1)
1) (School of Science, Wuhan University of Technology, Wuhan 430070)
2) (School of Automation, Huazhong University of Science and Technology, Wuhan 430074)
Abstract: Local binary pattern (LBP) has limitation in extracting texture feature and cannot effectively depict the
edge and direction information, thus a new method is proposed, called layered fusion with LBP and histogram of
oriented gradients (HOG) features. First, LBP operator is adopted to extract the layered texture spectrum feature
of an image, and then the edge features of the original image are extracted by using HOG operator, as well as the
layered HOG features which are based on the layered LBP. Finally, the layered LBP features with these two
different HOG edge features are fused to generate two different layered fusion features. The experiments are
implemented on ORL, Yale, GT face databases paring fifteen algorithms, which show that the layered
fusion features generated by the fusion method of this paper perform much better than the traditional
dimension-reduced algorithms, single LBP and single HOG. The corresponding recognition rates of the proposed
method are significantly improved, of which the best are 99