Discrete Spherical Harris Corner Detector

This paper proposes a method of detecting Harris corners from spherical retina image, which is assumed to be captured from a spherical retina camera. In a spherical retina camera, the photodetectors are supposed to be arranged on a sphere. Therefore, a spherical retina image is a discrete spherical image, which is obtained by tessellating a sphere into cells corresponding to pixels. In this paper, the gradient of each pixels is computed as a 3D vector represented at orthogonal coordinate system to avoid a pole problem [13]. The Harris corners are detected based on the 3D gradient vectors by modifying the conventional Harris corner detector for 2D planar image. The effectiveness of the proposed method is shown in the experimental results.