Elliptical Object Detection by a Modified RANSAC with Sampling Constraint from Boundary Curves' Clustering
Elliptical Object Detection by a Modified RANSAC with Sampling Constraint from Boundary Curves' Clustering
复制标题
DOI：
10.1587/transinf.e93.d.611
复制
发表时间：
2010-03
期刊：
影响因子：
--
通讯作者：
Yingdi Xie;J. Ohya
中科院分区：
文献类型：
--
作者：
Yingdi Xie;J. Ohya
本文提出了一种从图像中检测椭圆的方法，尽管 (1) 椭圆内有多种颜色，(2) 部分遮挡的椭圆边界，(3) 椭圆的噪声、局部变形边界，(4) 图像中存在除椭圆之外的多个对象，以及 (5) (1) 到 (4) 的组合。边缘检测得到边界曲线后，利用边界曲线中各像素的边缘方向的一阶差分曲线，通过分段重连接方法得到边界簇。然后，改进的 RANSAC 通过从边界簇中随机选择五个像素来检测椭圆，其中重叠的椭圆被合并。使用合成图像和真实图像的实验结果证明了该方法的有效性，并与众所周知的传统方法随机霍夫变换进行比较。
This paper proposes a method for detecting ellipses from an image despite (1) multiple colors within the ellipses, (2) partially occluded ellipses' boundaries, (3) noisy, locally deformed boundaries of ellipses, (4) presence of multiple objects other than the ellipses in the image, and (5) combinations of (1) through (4). After boundary curves are obtained by edge detection, by utilizing the first-order difference curves of the edge orientation of each pixel in the boundary curves, a segment-reconnect method obtains boundary clusters. Then, a modified RANSAC detects ellipses by choosing five pixels randomly from the boundary clusters, where overlapped ellipses are merged. Experimental results using synthesized images and real images demonstrate the effectiveness of the proposed method together with comparison with the Randomized Hough Transform, a well-known conventional method.