Ray-resampling geometric distortion correction for imaging systems

This Letter presents a ray-resampling method for direct geometric distortion correction based on densely calibrated pixel-to-ray mappings. Instead of relying on system-specific distortion models, calibrated rays are intersected with an object-space reference plane and resampled onto a regular grid, allowing the same correction workflow to accommodate substantially different imaging geometries. The method is further extended to light-field imaging through sub-aperture ray reconstruction and resampling. Experiments on wide-angle, fisheye, and non-central catadioptric systems achieve overall linearity errors of 0.0202, 0.0337, and 0.0557 px, respectively. In structured light-field 3D reconstruction, the standard deviation of plane-fitting error decreases from 1.8386 mm to 0.1709 mm, demonstrating accurate geometric correction and improved downstream 3D measurement.