摘要： 提出了一种多回路平面曲线的分割算法。算法以直线段和二次曲线段作为拟合的基本单元，通过边缘跟踪把多回路的平面曲线分割成曲线段或单回路的封闭曲线，然后在曲线曲率不连续处继续分割。针对一次分割时可能会产生的迷向问题，提出了基于对偶原理的线段合并技术。该算法的结果能简单有效地描述场景中的物体，使高级视觉任务更简单。实验结果表明算法能取得较好的效果。

关键词: 分割；对偶；多回路；平面曲线；合并

Abstract: Segmentation of digitized planar curves is very important in early image processing, which allows a meaningful and compact object description for higher-level vision processing. The ability of describing contours concisely, especially those corresponding to boundaries of object representation is necessary in task such as pattern recognition, shape matching, and geometric inspection. This paper proposes an algorithm, which segments a multi loop planar curve into straight lines and conics. In the algorithm, a multi loop curve is first decomposed into single curves by edge following, and then segments them into simple primitives, such as straight lines or conics according to curvature changes. On the basis of these, fitting technique is used to get the parameters of primitives. Finally, some of these primitives can be combined according to the parameters and end-points. Experiments are performed with two images.

Key words: Segmentation; Dual; Multi loop; Planar curve; Merging