节点文献
基于精化曲线拓扑边长的角点增强检测
Enhanced Corner Detection Based on the Topology Boundary of Refined Digital Curves
【摘要】 现有的直接角点检测方法难以处理多细节的角点检测问题。为此,本文提出一种角点增强策略:将直接方法所求出的角点强度评估值看成是实际强度值的平滑结果,通过选取适当的平滑函数,如Gauss函数,可以解出实际角点强度值的一个近似。这一策略的使用能够使得真角点容易被检测,而伪角点被有效去除。相应地,得到一种角点检测方法。该方法基于精化曲线的拓扑边长来检测角点,使得检测结果准确可靠,并具有旋转不变性。
【Abstract】 It is difficult for the existing direct corner detectors to perform well on curves with multiple --size features. In this paper a corner enhancement strategy is introduced. Since corner significance value computed by the direct detectors can be viewed as a smoothing result of the real significance, we can get an approximation of the real significance by choosing a reasonable smoothing function, such as Gaussian. With this strategy, the real corners are easily detected and the false corners are efficiently removed. Correspondingly, a new corner detector based on the Topology boundary of refined digital curves is developed. Its performance is robust to image noise and invariant to rotation transformations.
【Key words】 Digital Curve; Corner Detection; Computer Vision; Pattern Recognition;
- 【文献出处】 模式识别与人工智能 ,Pattern Recognition and Artificial Intelligence , 编辑部邮箱 ,2005年02期
- 【分类号】TP274.4
- 【被引频次】9
- 【下载频次】80