摘要： 为提高行人检测的识别率，提出一种基于改进型韦伯局部描述子(WLD)和局部二元模式(LBP)的特征融合方法进行行人检测。对图像进行二维离散Haar小波变换得到4个不同频率的子图像，对其中1个低频部分提取WLD特征，对3个高频部分提取LBP特征，并将各个子图像的特征串接为1个向量，得到WLD-LBP特征。在INRIA Person数据集上利用SVM作为分类器进行测试，实验结果表明，与单独WLD特征、梯度方向直方图(HOG)特征、PHOG特征以及HOG-LBP特征融合方法相比，该方法的识别率最高，可达98.1%，并且对光照和噪声也有较好的鲁棒性。
Abstract: This paper presents a feature fusion method(WLD-LBP) based on an improved Weber Local Descriptor(WLD) and Local Binary Pattern(LBP) through a two-dimensional discrete haar wavelet transform. The algorithm starts with a two-dimensional discrete haar wavelet for the image so as to obtain the subimages of four different frequencies. Making full use of the WLD and LBP, we extract the WLD characteristics of the low frequency part, and LBP features of the other three high-frequency portion, and then a vector consisted with the characteristics of the image is produced which we called WLD-LBP characteristics. Five groups of test experiments were conducted on INRIA Person databases using SVM as classifier,comparing with the characristics of WLD, Histogam of Oriented Gradient(HOG), PHOG and feature fusion of HOG-LBP,respectively. The results demonstrate the effectiveness with the highest recofnition rate up to 98.1% and robustness to illumination and noise of the proposed method.
2-D discrete wavelet transform,
Weber Local Descriptor(WLD),
Local Binary Pattern(LBP)