Deblurring method based on spatially varying PSF estimation

Abstract Time delay integration (TDI) cameras improve the signal-to-noise ratio through multi-stage charge accumulation, but line-rate synchronization errors cause row-dependent blur and geometric distortion. This paper proposes an encoder-guided non-blind restoration method that converts motion-stage displacement trajectories into spatially varying PSFs. Within each accumulation window, residual displacement is decomposed into a mean shift for distortion compensation and a centered diffusion component for PSF construction, with effective support constrained by the root-mean-square width of the centered offsets. The resulting PSFs are incorporated into overlapping sliding-strip Richardson–Lucy restoration. The method requires neither additional motion sensors nor TDI sensor modification. Simulations and physical experiments show improved edge and texture recovery. Compared with three baseline methods, the proposed method achieves higher PSNR at all tested blur levels and higher SSIM in most cases, with clearer advantages under mild-to-moderate blur. These findings demonstrate that encoder-guided spatially varying restoration effectively alleviates synchronization-error-induced degradation and reduces dependence on strict line-rate synchronization.