一种基于Harris-Laplace算法的亚像素角点检测方法

为了解决Harris-Laplace检测算法的角点坐标偏移与像素级角点的问题,提出了基于Harris-Laplace算法的亚像素角点检测方法。该方法首先用原始图像与高斯函数进行卷积生成多尺度空间,在原始图像和多尺度空间图像上各自提取Harris-Laplace角点;然后以多尺度空间角点为中心向原始图像投影,统计原始图像上投影区域内的角点形成角点集群,并结合多尺度空间角点响应值对集群角点进行筛选;最后采用位置(坐标)加权平均法确定角点的精确坐标。实验结果表明,该方法能够提供稳定抗噪、尺度不变的亚像素精度角点。

A Sub-Pixel Corner Detection Method Based on Harris-Laplace Algorithm

In order to solve the problem of the drifting of pixel position and the pixel precision of Harris-Laplace detection algorithm,a sub-pixel corner detection method based on Harris-Laplace algorithm is proposed.Firstly,the multi-scale space is generated by Guassian convolution with the original image,the Harris-Laplace corners are extracted both in the original image and multi-scale images.Then,the corners clusters are acquired by projection from scale space images to the original image,and the comers in clusters are sifted by the response value of the multi-scale space corners.Finally,the exact coordinates of corners are calculated by the coordinate-weighted average.Experimental results indicate that the proposed method can provide scale invariant sub-pixel corner points with the anti-noise stability.