LBP和HOG的分层特征融合的人脸识别

Fusion with Layered Features of LBP and HOG for Face Recognition

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%,99.5%和99.14%.

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 by comparing 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%, 99.5% and 99.14 %, respectively.