分层结构直方图及其应用

摘要

针对传统直方图特征存在的特征分辨能力不强的问题,提出了一种图像分层结构直方图.该特征首先将图像按照亮度幅值的大小进行分层,然后根据预先设计的结构图元对所有的分层图进行基于结构图元的直方图统计,最后将所得的直方图进行集成得到最终的分层结构直方图.以图像匹配与视觉跟踪为例,对分层结构直方图的应用进行了大量的仿真实验,结果表明该特征较参考特征具有更强的特征分辨性能和局部描述能力.利用分层结构直方图进行图像匹配得到的相似性度量图单峰特性更加明显,能够显著降低视觉跟踪算法的跟踪误差.%Aimed to the poor discrimination of the traditional histogram feature,a Hierarchical Structure Histogram (HSH) is proposed in this paper.Firstly,it transforms the target image into hierarchical sub images according to the intensity value.Then,it calculates the histograms based on the structure image element for each hierarchical sub image according to the predefined structure image element.Finally,it integrates the histograms to obtain the final hierarchical structure feature.Take image matching and visual tracking as two representative exanples,a lot of experiments were carried out based on theproposed feature.The experimental results indicate that the proposed feature owns better discriminative performance and local description performance than the referenced features.The similarity map obtained from the HSH-based image matching has a much more discriminative single-humped characteristic.The HSH feature can also dramatically decrease the tracking errors for visual tracking.