Robust corner detection in continuous space

Corner detection is important in image analysis and understanding, but most existing corner detectors are sensitive to image quality, lens radial distortion, and illumination. In this paper, we propose a corner detector for robust corner detection in continuous space. We use the open string theory to construct the continuous representation of an image. Defining a corner as the intersection of two or more curve edges or straight line edges, we design a corner response function for corner determination. In detail, for each integer point, we construct multiple grayscale-parallelograms by any two directed line segments of that point, and the corner response function is based on these grayscale-parallelograms. Finally, a point with a high response value is detected as a corner. Experimental results on conventional images, wide-angle images, and fisheye images show that the proposed method obtains state-of-the-art performance on conventional images and achieves superior performance on wide-angle images and fisheye images, even under weak lighting and low-quality conditions.