基于HOG特征描述的轮廓匹配算法  

Contour Matching Algorithm Based on HOG Feature Description

作　　者：童小彬 孟婥[1] 孙以泽[1] 张豪 TONG Xiaobin;MENG Chuo;SUN Yize;ZHANG Hao(College of Mechanical Engineering,Donghua University,Shanghai 201620,China)

机构地区：[1]东华大学机械工程学院,上海201620

出　　处：《东华大学学报（自然科学版）》2020年第5期787-792,共6页Journal of Donghua University(Natural Science)

基　　金：福建省百人计划资助项目(20180409);工信部智能制造新模式应用资助项目(20174682)。

摘　　要：针对现有轮廓匹配算法在轮廓缺失、轮廓噪声点、非线性光照下匹配效果差的问题,提出了基于方向梯度直方图(histogram of oriented gradient,HOG)特征描述的轮廓匹配算法。将轮廓曲线点离散化,计算所有离散点梯度并统计其在不同方向上的密度分布,以欧氏距离作为相似性度量,采用金字塔分层搜索策略进行搜索。统计所有离散点梯度信息得到梯度方向直方图,融合了轮廓的局部信息和全局信息,使得该算法对部分缺失、遮挡、边缘噪声具有较高的鲁棒性。以印花鞋面表面轮廓为试验对象,使用该算法进行匹配,结果表明,在复杂背景下,该算法对加入人为干扰的鞋面匹配效果较好,且采用金字塔分层搜索策略,匹配速度能满足实际应用的需要。Aiming at the poor matching effect of existing contour matching algorithm under missing contour,contour noise point and nonlinear illumination,a contour matching algorithm based on histogram of oriented gradient(HOG)feature description was proposed.The image contour curve points were discretized,all the discrete point gradients were calculated and their density distributions in different directions were counted.Euclidean distance was taken as the similarity measure and the pyramid hierarchical search strategy was adopted to search images.The gradient direction histogram obtained by all the discrete point gradient information was combined,and the contour local information and global information were fused,which made the algorithm more robust to image contour missing,occlusion and edge noise.The surface profile of printed upper was taken as the experimental object,and the algorithm was used for matching.The results show that the algorithm has a better matching effect on the uppers with the artificial interference in the complex background.The matching speed can meet the needs of practical applications by the pyramid hierarchical search strategy.

关 键 词：HOG特征 轮廓匹配 金字塔分层搜索 欧氏距离度量 印花鞋面 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]