Image Reconstruction
Overview
Image reconstruction asks how to infer an image, volume, or dynamic scene from sparse, noisy, blurred, distorted, or indirectly encoded measurements, as in MRI, CT, microscopy, and computational photography. Methods range from analytic and variational solvers such as filtered back-projection, compressed sensing, total variation, and plug-and-play to unrolled networks, implicit neural fields, Gaussian representations, and diffusion or flow-matching priors, often enforcing consistency with the measurements. Current work emphasizes self-supervision and test-time adaptation when paired ground truth is scarce, faster inference and joint sensing-reconstruction, and testing on real or shifted acquisition conditions. A central tension is that generative priors can restore perceptual detail from severe undersampling but may invent structure, so fidelity, uncertainty, and downstream task performance matter alongside pixel-level metrics.
Latest papers
297 papers33 papers in the last four weeks, up 3% on the four weeks before. 0.2% of all new papers.