摘要
针对形状上下文特征难以解决大规模样本的形状识别问题,提出一种利用角点典型形状上下文特征进行快速形状识别的方法.该方法仅以少数角点作为代表点生成直方图,对目标形状关键特征进行描述,通过减少匹配的特征数目降低了采样点匹配时间;在此基础上提出了局部约束匹配的方法,能够快速实现形状匹配并解决特征旋转不变性的问题,最终通过结合快速剪枝和精确匹配完成形状的识别.对形状数据进行仿真实验的结果证明,文中方法能够快速、有效地实现大规模数据的形状识别和检索.
In order to resolve computationally prohibitive problem of shape matching within a large database of shapes, corner representative shape context is proposed for shape recognition. With a small number of corners as representative points, the corresponding histograms provide descriptions for the key features of shapes, and the correspondences between sample points are determined rapidly by reducing the number of histograms. As locally constrained matching is further proposed, shape matching and rotation invariant can be effectively resolved, and shape recognition is achieved by a two stage approach, fast pruning and detailed matching. The proposed algorithm has been tested on databases of shapes, and the performances of our method are superior to many other methods.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2013年第2期215-220,共6页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61074096)
