Abstract

Accurate estimation of rotation angles between images is fundamental in image processing, with direct implications for computer vision and pattern recognition. This paper introduces a theoretically grounded method for rotation-angle estimation based on spatial statistical tools. The approach applies to spatial processes on both circular and rectangular grids and exploits the rotation-invariance property of the angular cross-variogram. By evaluating images over systematically rotated grids, the framework yields an estimator of the rotation angle between an original image and its rotated counterpart. The method's performance is assessed through simulations on circular and rectangular lattices, as well as through two real-image applications. Results show that the estimator performs reliably across various image textures, including in the presence of noise or contamination. This contribution provides a principled statistical foundation for rotation-angle estimation, with broad potential for image analysis and spatial statistics applications.