基于各向异性高斯核方向导数滤波器的图像轮廓检测

Image contour detection based on anisotropic Gaussian kernel directional derivatives filter

针对图像轮廓检测中存在大量毛刺和虚假轮廓的问题,提出一种基于各向异性高斯核方向导数滤波器和图论方法的图像轮廓检测算法。利用各向异性高斯核方向导数滤波器提取图像中各个方向的变化信息,使用阈值化方法实现图像中轮廓的初始检测。将初始检测的轮廓转换成加权无向图表示,利用图论中提取最短路径的方法递归地将轮廓的无向图转换成最短路径的集合。最后,使用阈值化方法消除初始轮廓检测结果中的毛刺和虚假轮廓。将检测结果与现在广泛使用的Canny检测结果相比,该方法基本消除了真实轮廓附近的毛刺,大大减少了虚假轮廓,获得了更好的检测性能。

Aiming at problem that there are a lot of burrs and false contours in image contour detection,a contour detection algorithm based on anisotropic Gaussian kernal directional derivatives（ANDDs） filter and the graph theory is proposed.ANDDs filter is used to extract change information of image in all orientation.Based on which the primitive detection of contour in image is realized utilizing threshold procedure.Transform primitive detected contours into the representation of undirected and weighted graph,then the algorithm extracting the shortest path of an undirected and weighted graph is used iteratively to subdivide the graph into a set of path.Finally,threshold method is used to eliminate the burrs and false contours in primitive contour detection.Test results show that the proposed method basicly eliminate burrs around real contour and greatly reduces false contours,and obtain better detecting performance,compared with Canny method.