基于曲率尺度空间的角点检测教程文件.ppt

Curvature Scale Space Based Image Corner Detection;Corner detection is an important task in various machine vision and image processing systems. Applications include motion tracking, object recognition, and stereo matching. It is a fundamental problem and several different algorithms have been proposed. 角点是图像亮度发生剧烈变化或图像边缘曲线上曲率极大值的点。在实际地物中，存在大量道路和房屋的拐角“┐”、道路十字交岔口“X”、“+”等，这些地物体现在图像中，就是图像中得角点。 （书P45） 角点有两个版本的定义：一是角点可以是两个边缘的交点；二是角点是邻域内具有两个主方向的特征点。从20世纪70年代至今，许多学者对交点测试投入了大量的研究，因此也产生了很多角点检测算法。 这些算法大致可以分为三类： 基于模板的角点测试方法；（基于二值图像的角点检测） 基于边缘特征的角点测试方法；（基于轮廓曲线的角点检测） 基于亮度变化的角点测试方法。（基于灰度图像的角点检测） （书P37）;Propose the following criteria: All the true corners should be detected Corner points should be well localized No false corners should be detected Corner detector should be robust to noise Corner detector should be efficient; 曲线的曲率（curvature）就是针对曲线上某个点的切线方向角对弧长的转动率，通过微分来定义，表明曲线偏移直线的程度。数学上表明曲线在某一点的弯曲程度的数值。曲率越大，表示曲线的弯曲程度越大。曲率的倒数就是曲率半径。 例如在曲线CD上点A和临近一点A’各做一条切线，A和A’之间的弧长为ΔS，两条切线夹角为α，则曲线CD在A点的曲率为 ：;The curvature scale space technique 曲率尺度空间技术;In order to find curvature zero-crossings or extrema from evolved versions of the input curve, one needs to compute curvature accurately and directly on an evolved version ?? of that curve. Curvature on ?? is given by: 曲率演化形式定义 where The function defined implicitly by is the CSS image of ??. 其中， 为曲线 的曲率值， 和 分别为对应 的关于 一阶和二阶导数。 当 时，得到曲线的曲率尺度空间图像。;Figure 1: Evolution of Africa; As the contour evolves, the actual locations of the corners change. If the detection is achieved at a large scale the localization of the corners will be poor. To overcome this problem tracking is introduced in the detection. The corners are located at a high scale , this assures that the corner detection is not affected by noise. Then sigma is reduced and the same corner points are examined at lower scales. As a result, location of corners may be updated. This is continued until