Two Sub-Pixel Corner Detection Methods

This article presents two sub-pixel corner detection methodologies: surface fitting and grayscale gradient analysis. The surface fitting method typically employs quadratic polynomial interpolation around initial corner estimates to achieve sub-pixel accuracy through least-squares optimization. The grayscale gradient approach calculates gradient vector intersections using local intensity variations for precision enhancement. Both algorithms generate precise coordinate outputs with superior accuracy compared to standard pixel-level detectors. The main function serves as the primary executable, orchestrating image input, corner initialization, and sub-pixel refinement processes. Key implementation steps include: - Gaussian smoothing for noise reduction - Harris or Shi-Tomasi corner pre-detection - Sub-pixel refinement using cv2.cornerSubPix() in OpenCV - Coordinate validation and output formatting These methods enable more accurate corner localization in digital images, providing reliable foundations for subsequent computer vision tasks such as camera calibration, 3D reconstruction, and image registration. The grayscale gradient method particularly excels in textured regions, while surface fitting demonstrates robustness under varying illumination conditions.