基于区域检测的多尺度Harris角点检测算法

Multi-scale Harris-corner detection algorithm based on region detection

针对传统Harris算法检测的角点遍布整个图像和对尺度变化较敏感的问题,本文提出将区域检测方法和多尺度Harris角点检测算法相结合,使检测到的角点数目更少,以提高后续的图像匹配重建的效率。首先,利用区域检测的算法构造图像显著图,采取腐蚀膨胀操作提取出目标区域作为候选的检测区域;其次,利用多尺度结合非极大值抑制的方法改进Harris算法,检测图像的角点并标记。仿真结果表明:本文方法能进一步提高角点检测的精确度和速度,同时在不改变任何参数的情况下,对于图像旋转能够减小角点提取的差异,增强算法的多尺度性。

In this paper,we address the problems associated with the Harris-corner detection algorithm,in which corners are extracted from the entire image and there is sensitivity to scale change. We combine region detection and the multi-scale Harris-corner detection algorithm,such that fewer corners are detected and more efficient subsequent image matching and reconstruction is achieved. First,we use the region detection algorithm to construct an image saliency map,and use the erosion and dilation operations to extract the target region as a candidate detection region. Next,multi-scale method combines with non-maxima suppression method is used to improve Harris algorithm,to detect the corners and mark. Simulation results show that the method achieves better accuracy and speed in corner detection. Furthermore,without changing any parameters,this algorithm reduces the differences due to the extracted corners and enhances the multi-scale of the algorithm for image rotation.