基于图像分块的多尺度Harris角点检测方法

Harris角点检测是一种经典的角点检测算法，在现实中应用广泛，但不具有尺度变化特性。为了改变其单一尺度的特性，使得角点提取更加精确和有效，将多尺度的概念和图像分块方法引入到Harris算法中，在多个尺度下对角点进行提取。将每个尺度上的角点响应值的本地最大值作为该尺度上的候选角点，并同时对图像进行分块；最后，沿小尺度到大尺度方向判断候选角点是否是真实角点，剔除伪角点，使得角点检测更加精确。通过对比实验，新算法明显地提高了图像角点的检测性能。

Multi-scale Harris corner detection based on image block

Harris corner detection is a classical algorithm and is widely used nowadays, but it does not have the invariant property. For modifying its single-scale and making corner detection more accurate and valid, the conception of multi-scale-space and image block method were introduced into the Harris algorithm in this paper. Corner detection was used to corner extraction by multi-scale-space. Under each scale, the maximum response of local corner was regarded as a candidate corner point, while the image was segmented. Finally, along the direction of small scale to large scale to judge whether the candidate corner was true, the false corner was eliminated, thus the corner detection was more accurate. By comparison test, the new algorithm significantly improves the image corner detection performance.