基于平行四边形对角线理论的角点检测算法

Corner Detection Algorithm Based on Parallelogram Diagonal Theory

角点检测是图像处理和计算机视觉领域的基本任务,角点响应函数构造的复杂性或者曲线进行多次平滑的操作会制约角点检测方案的检测效率。针对这一问题,提出一种利用平行四边形对角线之比快速估计曲率的角点检测算法(fast corner detecion based on the ratio of parallelogram diagonals,FRPD)。首先,利用Canny边缘检测器提取边缘轮廓线,通过各向异性高斯方向导数滤波器对边缘线进行平滑;其次,利用提出的角点响应函数估计曲线上每个像素点的离散曲率,将曲率值大于设定阈值的像素点作为候选角;最后,对候选角进行非极大值抑制,删除弱角点和伪角点,保留精确的角点集。实验结果表明:与现有五种基于轮廓的角点检测CTAR、SODC、GCM、CPDA、F-CPDA算法相比,利用平行四边形对角线之比的角点检测FRPD算法不需要平方根运算,大大降低了计算复杂度,且该算法在相同的图像数据集下平均重复率最高,定位更加准确,具有优异的角点检测性能,角点检测速度约是CTAR的3倍,对噪声具有良好的鲁棒性。

Corner detection is one of the fundamental topics in image processing and computer vision. The complexity of the construction of the corner response function or the multiple smoothing of the curve often restricts detection efficiency of the corner detection scheme. Thus,a novel method for image corner detection based on the diagonal of a parallelogram to was proposed estimate the curvature value in this paper. Firstly,the Canny edge detector was used to extract each edge contour from the input image. Secondly,curves were smoothed by using anisotropic Gaussian directional derivative filter,the discrete curvature of each pixel on the curve were estimated according to the corner response function proposed in this paper. And then,non-maximum suppression was applied to the candidate corner sets. Finally,the refined corner sets were retained with unstable and false corners removed. Compared with the existing five contour-based corner detection algorithms,the proposed algorithm did not require square root operation. The extensive experiments showed that the developed method could give the highest average repeatability and lowest localization error than the other five detectors,while the corner detection speed was about 3 times that of CTAR. The results showed that the corner detection algorithm using the ratio of parallelogram diagonals( FRPD) not only had excellent corner detection performance,but also greatly reduced the computational complexity,and has a good noise robustness.