Accelerated image resampling for geometry correction

Abstract

This paper introduces a fast algorithm intended for corrections of image distortions caused by lenses or similar devices in real time. Not only the algorithm itself is introduced, but also the paper focuses on its acceleration with the focus on programmable hardware. The presented algorithm is designed to correct small geometrical distortions, where the displacement of the location is only few pixels. However, the resampling is done with high sub-pixel precision. Description of the geometry of the distortions is done through a square or rectangular mesh, where simple bilinear interpolation within each node is exploited to calculate sub-pixel location of the samples. The paper describes the algorithm, its features, specifics the important for implementation, data formats, implementation in field programmable gate arrays, and summarizes the achieved results.