针对大规模三维重建应用中多幅无序图像的分组及其有序化问题,提出一种鲁棒的无序图像分组方法.首先,对大量无序图像提取SIFT(scale invariant feature transform)特征,接着采用DBH (dichotomy bassed hash)算法对图像特征集合进行快速匹配,最后应用一种新的基于图像内容的图像相似度度量准则,将无序图像分组并采用视图生成树拓扑化组内图像.实验结果表明,该算法能快速有效地对无序图像分类并拓扑化.%Aim. The introduction of the full paper reviews some papers in the open literature and then proposes what we believe to be a novel and relatively more effective grouping and organization algorithm, which is explained in section 1. Section 1 describes the feature similarities of any two images to judge their correlation; subsection 1.1 gives a four-step procedure for the dichotomy-based Hash (DBH) feature matching algorithm; subsection 1.2 presents the grouping strategy for SIFT (scale invariant feature transform) matching; its core is that, on the basis of the complicated and correlated image similarity measure proposed by Yao et al in Ref. 4, we propose a novel image content-based similarity measure, which is given in eq. (2), to categorize the unordered images into their corresponding groups and further produce their view-spanning trees. To verify the effectiveness of our grouping strategy,section 2 did experiments on two sets of challenging image data; the experimental results, given in Figs. 2, 3 and 4 and Table 1, and their analysis show preliminarily that our algorithm can effectively group and organize a large number of unordered images, while not requiring the modification by any other methods.
展开▼
