Computational Imaging

Computational imaging studies how to acquire and reconstruct images when optics or sensors encode scenes indirectly, as in CT and MRI, lensless and single-pixel cameras, hyperspectral systems, event cameras, and photon-limited imaging. Methods range from physics-based inverse optimization and compressed sensing to deep unfolding, self-supervised learning, and diffusion or flow-based generative priors; many also co-design illumination, masks, exposure, or sensor layouts with reconstruction. Researchers assess fidelity, uncertainty, speed, and robustness in settings such as sparse-view tomography, demosaicing, depth estimation, and non-line-of-sight imaging, using both simulations and real hardware. A central tension is that learned priors can fill gaps in measurements but may hallucinate when the forward model or calibration is wrong, while accurate physics, reliable uncertainty estimates, and efficient inference remain challenging.