Enhancing lensless imaging via Explicit Learning of Model Mismatch

Yin Xiangjun, Miao Sun, Zhengyuan Zhang, Pengfei Qi, Huihui Yue

Keywords: Lensless Imaging; Maximum a Posteriori;Model Mismatch Error;Image Reconstruction

Abstract:

Emerging lensless imaging techniques hold promise for miniaturized cameras, but their effectiveness is constrained by challenges like model mismatch from the point spread function (PSF), which undermines reconstruction methods dependent on accurate PSF modeling. To address this issue, we propose a joint Maximum a Posteriori (MAP) approach to simultaneously estimate model mismatch error (${\rm M^{2}}$E) and reconstruct high-resolution images from lensless imaging measurements. Specifically, we propose an explicit latent space representation for ${\rm M^{2}}$E to improve robustness against PSF inaccuracies. Additionally, we develop a multi-stage reconstruction network by unfolding the joint MAP estimator with a learned Laplacian Scale Mixture (LSM) prior and ${\rm M^{2}}$E representation (${\rm M^{2}}$ER) through end-to-end optimization. Extensive experiments show that our method surpasses current state-of-the-art methods.

Primary Area: interpretability and explainable AI

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6256