Scene-aware virtual lens group for achromatic extended depth-of-field imaging.

Abstract

Achieving extended depth-of-field imaging while maintaining high color fidelity remains challenging for traditional optical systems due to intrinsic trade-offs. Aside from the need for bulky optical elements, current algorithms also struggle with wavelength-dependent blur, which worsens at larger imaging depths due to nonlinear defocus and a reduced field of view. To overcome these limitations, we propose a physics-based optical-field manipulation pipeline that achieves achromatic extended depth-of-field (AEDOF) imaging using a single off-the-shelf refractive lens. The distorted light field produced by the single lens is reconstructed to enhance the utilization of physical information, bridging the gap between incoherent and coherent imaging. Multiple virtual lenses are dynamically positioned according to aberration-depth reliance, directly correcting wavefront distortions rather than relying on conventional slice-based pixel-wise processing, thus transforming high-fidelity large space-bandwidth product (SBP) imaging. Tested on various scenes, our methodology achieves state-of-the-art imaging performance, with a peak signal-to-noise ratio of 30.1322 dB. The proposed scene-aware differentiable AEDOF computational imaging approach provides a feasible and configurable solution for applications such as ultra-compact telescopes and precision biomedical diagnostics.