摘要
Contour grouping is the precondition for high-level visual tasks such as region-of-interest detection and object recognition. Those methods adopting local geometric features and simple photometric attributes always lead to unreliable result or bad robustness. In this paper, we propose a grouping method using multiresolution analysis and one-dimensional manifold. By using multiresolution analysis, grouping seeds and initial clusters are detected as precondition for describing topological structure. Then one-dimensional manifold is applied to discovering intrinsic order of topological structure, which corresponds to coordinates of edges in a closed contour. These two improvements enhance our method's robustness relative to those using local features or linear combinations of them. Experiments on different classes of real images show competence of our method.
- 出版日期2007-1
- 单位中国传媒大学; 北京交通大学