An efficient and robust corner detection algorithm

Corner detection has been shown to be extremely useful in many computer vision applications. In this paper, an improved SUSAN corner detector is proposed and its performance is compared with Harris and SUSAN corner detection. The method adopts an adaptive multi-threshold strategy based on local brightness rather than one threshold for the whole image, and divides the circular mask area of SUSAN into two or more parts. Next, the number of pixels in the part where the nucleus locates is calculated. If the number is less than half the total pixels of the circular mask, the nucleus becomes a corner candidate. The exact positions of image corners are those candidates with local minimum numbers. In order to improve the computational efficiency of the algorithm, we limit the search space for corner candidates to those pixels whose intensity gradient magnitudes by Sobel operator are higher than a threshold. Experiments have demonstrated that our corner detector is accurate and efficient.