基于k_d曲率的自适应支撑邻域角点检测

k_d Curvature-Based Adaptive Support for Neighborhood Corner Detection

根据kd曲率计算方法具有比传统方法简捷的特点,提出了基于kd曲率自适应支撑邻域的角点检测方法.首先,用Canny算子提取初步轮廓曲线,然后进行融合填补,再筛选出高质量的轮廓曲线,并对曲线进行高斯核平滑.从算法鲁棒性考虑,提出一种新的思想,即寻找一个可以确定一条曲线总体支撑邻域的参数,使其随着仿射变换在噪声干扰下发生有规律的变化,从而实现自适应支撑邻域的角点检测,并用自适应阈值和非极大值抑制来排除伪角点和弱角点,最后提取精确的角点.通过实验与Harris、He&Yang、CPDA、KD、ANDD等算法对比,该算法的定位误差和错误率较低,而平均重复率明显较高,具有更好的角点检测性能.

kd curvature calculation method is simpler and faster than traditional curvature calculation method. In this paper a corner detection method based on kd curvature adaptive support neighborhood is proposed. Firstly, the initial contour curve is extracted by Canny operator, fused and filled, then the high quality contour curve is selected, and the curve is smoothed by Gaussian kernel. From the point of view of algorithm robustness, this algorithm proposes a new idea, which is to find a parameter that can determine the total support neighborhood of a curve, and this parameter changes regularly with affine transformation and noise interference, so as to realize the corner detection of the adaptive support neighborhood, and uses adaptive threshold and non-polar. Large value suppression is used to exclude false corners and weak corners. Finally, the precise corner points are extracted. Compared with Harris, He ＆ Yang, CPDA, KD and ANDD, the positioning error, error rate and average repetition rate are improved by 25%. Therefore, the algorithm has better corner detection performance.