摘要
针对多圆检测问题,提出了一种基于RHT的改进算法PHT3(3-point Hough transformation)。对整幅图像特征点按连续性进行点集归类,同时计算有效点的梯度方向信息;按照一定的取点规则在同一点集中取3点,得到候选圆的圆心参数;依据所求圆心参数以及梯度信息判定选取3点的有效性,以降低Hough变换的无效累积。针对常规确定圆半径精度有限的缺陷,提出利用点集并结合候选半径的均方差来获得亚像素半径,同时解决了同心圆半径的检测问题。与RHT算法进行对比检测,结果表明:PHT3算法检测时间为RHT算法检测的1/6,且无效累积更小,同时保留了Hough变换对局部信息缺损不敏感和对随机噪声鲁棒性强的特点。
In this paper, an improved arithmetic, named PHT3 （3-point Hough transformation）, for multi-circle detection was developed based on RHT （randomized Hough transformation）. At first, the PHT3 provides a special method, which classifies all the characteristic points of an image based on the continuity points and records the valid points＇ gradient direction information. And then a candidate circle center point is determined based on three points in one class selected by the certain rule. Finally according the circle center and the corresponding points＇ gradient direction information the point-group＇s validity is verified, which can reduce the useless accumulation. Aiming at the defect of low precision of the calculated radius using a general method, a new method based on the information of candidate radius and their mean variances was proposed to obtain the sub-pixel radius that can also be used to detect the concentric circles＇ radius. Through the comparison between the RHT and PTH3, it is concluded that PHT3＇s processing time is just one sixth of the RHT, useless accumulation is greatly reduced and the features of insensitive to partial message damage and strong robustness to random noises of the Hough transformation are preserved.
出处
《中国农业大学学报》
CAS
CSCD
北大核心
2008年第4期121-125,共5页
Journal of China Agricultural University
基金
国家“十一五”科技支撑项目课题(2006BAD28B03)
