The interesting point is important features in images, which is frequently used for image registration, scene analysis, object tracking. Many algorithms for detecting the interesting points have been developed up to now. Among them, the Harris interesting point detector is proved most stable against rotation and noise, while it is also very sensitive to scale change. In order to solve this problem, a new multi-scale Harris interesting point detection algorithm is proposed in this paper. The algorithm consists of two main stages. In stage one, we build a multi-scale representation for Harris measure and find candidate interesting points at each scale level. In stage two, a novel measure is defined to measure the stability of each point. Based on the measure, the final interesting point is selected from candidates. The experiments on both optical and sar data sets have shown that, compared with stand Harris interesting point detector and some other multi-scale interesting point detectors, our algorithm is more robust.
