During the corner feature extraction of unmanned aerial vehicle (UAV) images of high mountains and dense forests, the number of corner features was small by using traditional methods, thus a self-adaption extraction method being suggested, in which the whole process could be divided into two parts. The first part is primary corner feature extraction which determines the candidate corner feature from images with the Lv-Yan operator. While the second part is accurate corner feature extraction, which accurately extractes corner feature from the candidate corner feature with improved SUSAN algorithm. The experimental results have proved the feasibility of the proposed approach, which could extract greater number of corner feature points and run faster compared with traditional methods, as well as adapt to the corner feature extraction of UAV images in high mountains and dense forests.