Study of Harris scale invariant keypoint detector

程邦胜,唐孝威
DOI: https://doi.org/10.3785/j.issn.1008-973X.2009.05.013
2009-01-01
Abstract:Previous works demonstrated that Harris scale invariant detector with particular parameters cannot extract enough stable keypoints from images, and this led to the conclusion that Harris scale invariant detector was unstable and invalid for feature detection. In the process of constructing the Harris cornerness scale space, there are some parameters influencing the number of local maxima in the scale space and determining the number of keypoints that can be extracted by Harris scale invariant detector. This work systematically investigated this parameter space and found that the ratio between integrate scale and differentiate scale had deterministic influence on the number of keypoints that could be extracted by Harris scale invariant detector. When decreasing this ratio, the detected keypoint number increases rapidly; when this ratio equals to 0.8, the detected keypoint number is comparable to that detected by DoG detector. This valid parameter interval greatly increases the keypoint number detected by Harris-Laplace detector. When comparing with DoG detector and Harris-Laplace detector, Harris scale invariant detector obtains the best performance. This result corrects the previous conclusion about the instability of Harris scale invariant keypoint detector.
What problem does this paper attempt to address?