基于图像轮廓曲线的拐点检测算法

拐点是图像很重要的特征,包含图像的主要结构信息,拐点检测对图像配准、目标识别和三维重建等任务的处理都有着非常重要的作用。从研究轮廓上点的曲率出发,针对传统基于轮廓的拐点检测算法对噪声和局部变化敏感而造成检测结果不稳定的问题,提出一种间接反映轮廓曲率特性的方法,记为APTD(Accumulation of Point to Tangent Distance)。使用高斯函数对轮廓曲线进行平滑处理,轮廓上点的曲率越大则该点附近的点到该点处切线的距离也越大,根据这一思想,将轮廓上点附近支撑域内的点到该点所对应切线相对距离的累加和作为拐点的判别函数,从而实现轮廓拐点的检测。经由数学推导表明所提出的算法具有合理性和可行性,对比实验分析表明该算法精度高、运算量小、定位准确。

Corner Detection Algorithm Based on Image Contour Curve

Corner is a very important feature of the image,which contains the main structural information of the image.The corner detection plays a very important role in the processing of image registration,target recognition and three-dimensional reconstruction.Starting from the study of curvature of points on the contour,a new estimation method,APTD (Accumulation of Point to Tangent Distance),which indirectly reflects the curvature characteristics of the contour,is proposed to solve the problem of unstable detection results caused by the traditional contour corner detection algorithm being sensitive to noise and local changes.The contour curve is smoothed by the Gaussian function,then for a point on the contour,the larger the curvature of the point,the greater the distance between the point near the point to the tangent at the point.According to this idea,the sum of the relative distance between the point in the support area near the point on the contour and the tangent line corresponding to the point is used as the corner discriminant function,so as to realize the detection of the corner on the contour.The mathematical derivation shows that the algorithm is reasonable and feasible,and the comparative experimental results show that it has high precision,small computation and accurate positioning.