Abstract: Corner is a very important feature in the fields of image registration and target recognition. And it is a hot issue how to extract corners that are invariant to geometry transform and intensity change. Algorithms that use intensity directly to extract corners are robust, but poor in localization. Algorithms that use the information of edges to extract corners are outstanding in localization and invariant to geometry transform, but poor in robustness. If making use of these two kinds of information, an algorithm may extract corners with good property in every aspect. So we design an algorithm for corners extraction combining edge information and fractal signature. The algorithm is divided into three steps: firstly, extract edges of image and organize into groups. Secondly, compute the properties and distinct values of every point on the edge. Finally, extract satisfactory corners using nonmaximal suppress. The experiments show that the proposed algorithm for corner extraction can deal with the case of rotation and zoom of the image, and has good property in localization.