基于欧氏距离的拐点检测算法

Corner detection algorithm based on euclidean distance

拐点是数字图像中的一个重要信息载体,提出一种新的拐点检测算法,该算法并非寻找连续空间中曲率的离散近似计算方法,而是源于离散曲线的外观特征,推导出离散曲线上拐点处k个点对间欧氏距离平方和局部最小这一重要性质。基于该性质,本算法首先利用Freeman链码的性质过滤掉物体边界上明显不可能成为拐点的象素,然后在剩余的边界点中通过寻找该局部最小值定位出拐点。给出了本算法与四种著名拐点检测算法的对比实验。

Corners are important information carriers in computer vision. A new algorithm was presented here to detect corners on contour in digital image. This algorithm was not going to search another way to approximately calculate the curvature of points on curves,which was defined in continuous domain,but utilized the character of corners in digital nature that the square sum of k Euclidean distance between points pair centered at a corner is locally lowest. Derived from this character,the new algorithm detected corners in a two-pass manner. First pass was to filter the points on a curve that obviously can not be corners by using Freeman chain-code. Second pass was to detect the locations of local minima of the square sum of Euclidean distance. Tests comparing the new algorithm to four famous algorithms were given.