Deep learning-assisted metalens imaging over a wide depth of field

Miniaturized lenses with a large depth of field and high imaging quality are desirable for compact optical systems, as they eliminate the need for lens switching and repeated refocusing. Metalenses, composed of flat, subwavelength nanostructures, are well suited to this demand due to their ultra-thin profile and design flexibility. However, miniaturized metalenses typically require larger numerical apertures (NA), which lead to strong chromatic dispersion and resolution degradation. To address this limitation, we propose a Metalens Depth-of-Field Generative Adversarial Network tailored for restoring full-color images captured by a high-NA (0.447) millimeter-scale metalens. It achieves a 35% increase in peak signal-to-noise ratio and a 57.7% reduction in perceptual loss, while maintaining reconstruction quality across over 17.5 cm depth of field without additional training. This network provides a practical and scalable solution for enhancing image quality in miniaturized imaging systems.