Skip to Main Content
Corner that is invariant to geometry transform and illumination change is a very important feature for image registration and target recognition etc. Corners extracted using intensities directly are robust, but with poor property in localization. Corners extracted by edge information are invariant to geometry transform and outstanding in localization, but not robust. Therefore in this paper, an algorithm is designed for corner extraction, which combines fractal signature, edge information and multi-scale analysis. The proposed algorithm is divided into four steps: firstly, extract edges from the image; secondly, compute the properties and prominent values of every point on the edge; thirdly, extract satisfactory corners using non-maximal suppress; finally, compute the descriptor of extracted points. The experiment results show that the proposed algorithm is robust to scale, rotation and illumination, and has good property in localization.