Neue Publikationen in Ziel-Venues – 2026-10-13
Neue Werke: 1 · davon mit DE-Affiliation: 1 Venues: 60 · abgefragt ab publication_date 2026-09-13 · letzter Lauf: 2026-10-13T12:02:40
Optics Express (OE) — 1 neu · photonics
Reference-guided defocus estimation enables physics-guided image restoration in deep ultraviolet wafer microscopy ⚑ DE
- DOI: 10.1364/oe.616126
- Metadaten: Erschienen: 2026-10-08 · Vol. 34, Issue 21, S. 39815 · OpenAlex seit 2026-10-09
- DE-Institutionen: Institute for Advanced Study
- Autoren:
- Xiangji Guo — Institute for Advanced Study (DE)
- Yubo Wang — Hangzhou Dianzi University (CN)
- Ming Ming — Institute for Advanced Study (DE)
- Topics: Image Processing Techniques and Applications; Digital Holography and Microscopy; Advanced X-ray Imaging Techniques
- Keywords: defocus estimation; aberration correction
- Abstract: High numerical aperture deep ultraviolet (DUV) microscopy provides high spatial resolution, but its limited depth of field causes defocus degradation that weakens high-frequency information. Restoring such images is challenging because specimen structures are coupled with optical blur, while conventional restoration methods generally lack explicit constraints on the degradation process. Here, we propose a reference-guided defocus estimation and PSF-guided restoration framework for DUV wafer microscopy. A reference-guided differential-gated spatial–frequency estimation network (RDG-SFENet) estimates the defocus distance from the relative degradation between a defocused image and an in-focus reference that does not require strict registration, and the corresponding PSF is then obtained from the calibrated defocus relationship. A defocus PSF-guided data-consistency network (DPSF-DCNet) further incorporates the estimated PSF into Tikhonov-regularized pre-deconvolution and conditional feature modulation to constrain image restoration. RDG-SFENet achieves a defocus estimation MAE of 0.018 μm, while DPSF-DCNet reaches an average PSNR of 42.35 dB, exceeding the best-performing general restoration network by 2.93 dB. Defect experiments further show that the proposed method preserves defect morphology and contrast under defocus. These results highlight the potential of the proposed framework as a practical computational solution for extending the usable depth of field and enabling high-fidelity defect inspection in high-resolution DUV wafer microscopy.