Abstract:

Harris corner detection is a classical algorithm, but the detection result is closely related to selection of the Gaussian window length and the threshold value. Aiming at the problems, an improved algorithm for Harris corner detection is proposed. To begin with, scan the whole image with a lower threshold to extract the suspected corners. And then, a new circle mask is presented to evaluate the gray difference of pixels in the outside circle and the inside circle. The fake corners are eliminated by evaluating every suspected corner with the new circle mask and get the result. The proposed algorithm, avoiding derivative, is stable and robust. The experimental results of a series of images show the effectiveness and feasibility of the proposed method.