基于Radon变换的数字图像角点定位

Corner Localization in Digital Images Based on Radon Transform

定位精度是衡量特征点性能的重要指标之一,研究高定位精度的特征点检测方法对有定位精度要求的应用如三维重建、运动估计、目标跟踪与识别、图像配准等有重要意义。本文提出一种基于Radon变换的角点定位算法,该算法利用传统的边缘检测获得边缘二值图像,通过Radon变换估计二值图像中理想的直线和曲线,计算理想直线、曲线的交点作为潜在角点,剔除伪角点。模拟图像与真实图像实验表明,该方法对噪声有很好的鲁棒性,且能准确定位数字图像中的角点位置,定位精度优于现有的特征点检测方法。

Location accuracy is one of the key indexes of measuring the performance of feature point.. High location accuracy feature point detection has important implications for the applications such as 3 D reconstruction, motion estimation, target tracking and recognition, image registration, etc. A corner location algorithm based on Radon transform is proposed.Edge binary images are obtained with traditional method first, ideal straight lines and curves are extracted then, potential corner locations are calculated next, fake corners are rejected finally. Experimental results on both simulation and real images show that, this method performs better than the existing methods on interest point localization accuracy. Meanwhile, the strategy makes this method robust to image noises.