Centroid-guided localized super-oscillatory reconstruction enables large-field-of-view super-resolution imaging

Abstract

Conventional super-oscillatory imaging suffers from severe inter-object cross-talk and background contamination in large field-of-view scenes due to the extended sideband distribution of the point spread function. We propose a centroid-driven localized numerical super-oscillatory reconstruction framework for sparse, large field-of- view imaging. Object positions are automatically determined using connected-component analysis and spatial moment-based centroid estimation, enabling patch-wise processing without manual intervention. Each object is locally isolated, re-centered, and convolved with an engineered super-oscillatory point spread function, followed by a dual-windowing scheme to suppress boundary artifacts and confine the reconstruction within a predefined region of interest. The independently reconstructed patches are then reassembled via inverse translation and linear superposition. Simulation results demonstrate effective suppression of inter-object cross-talk and background artifacts, while preserving sub-diffraction structural details across an extended field of view.