一种分块的基于Harris角点检测的图像配准方法

传统的Harris角点检测选用全局的阈值并且不具有尺度不变性，对于较大的图像会导致检测的角点分布不均、错检等问题，为此提出一种新的基于多尺度的Harris角点检测的图像配准方法。首先将图像分块，并对其进行相应排序，根据局部阈值来提取Harris角点，然后根据图像特征点的最邻近和次邻近距离之比来确定初始匹配，最后利用特征点附近的灰度信息来实现进一步的配准。实验证明此方法使得图像配准精度和配准效率得到了极大地提高。

A Block Image Registration Method Based On Harris Corner Detection

For traditional Harris corner detection, threshold is a global value and has not the property of scale invariant, which cannot get a satisfied result in the image of corner detection, especially in larger images. A new image registration based on multi-scale Harris detector is represented on this paper. At first, it blocks the images , sorts correspondingly, and extracts Harris corner according to local threshold, then initially matches the images through the ratio between the distances of the nearest neighbor to the point and the second-nearest neighbor to the point. At last, image registration is carried out according to the gray information around the points. The experiment proved that this method makes the precision and efficiency of image registration improved greatly.