Calibration-preserving frequency-domain deblurring for optical sub-pixel metrology under composite degradation

Optical vision measurement in high-throughput industrial inspection must achieve sub-pixel positioning accuracy under severe time constraints. Rapid mechanical motion causes composite optical degradation, thereby introducing motion blur, defocus, and lens aberrations, attenuating high-frequency edge features and causing systematic positioning errors beyond precision tolerances. Existing computational imaging pipelines employ full-frame resampling that, if not explicitly corrected, introduces interpolation-induced geometric distortions that compromise the physical-to-pixel calibration mapping and the metrological traceability of restored outputs. This paper presents a calibration-preserving optical deblurring framework that maintains calibrated physical-to-pixel correspondence throughout the measurement pipeline. The core innovation is a phase-consistent frequency-domain restoration architecture where the network modulates only the magnitude spectrum while the original phase that encodes edge position information is retained in the frequency branch. Boundary-aware spatial refinement then selectively sharpens edge contours according to local gradient cues. A metrology-driven composite objective weights central pixels by morphological distance and enforces gradient fidelity. Validated on surface mount technology inspection, the framework achieves significant improvements in measurement accuracy and positional repeatability under real-time industrial constraints.