Multi-scale correlation construction for passive low-light imaging with a SPAD camera

Abstract

Recent advances have significantly improved single-photon avalanche diode (SPAD) imaging technology. However, most prior research has focused on active imaging using SPADs. To address the limiting application scope, we propose the multi-scale correlation construction method that enables low-light passive imaging using SPADs. The proposed method decomposes the reconstruction of binary photon streams into global, block, and pixel sub-problems, enhancing adaptability by not relying on temporal cues or deep learning hardware. Under low-light conditions (photons per pixel, PPP = 0.9), the proposed method achieves a root mean square error (RMSE) of 0.18. The proposed method demonstrates the ability to capture the primary objects of a scene with a few bit planes under the same photon-limited conditions. We refer to this capability as the “enrichment of main information under low light.” Quantitative evaluations using average gradient and information entropy metrics confirm the method's effectiveness in enhancing key structures and details while reducing computational overhead. Experimental results show that, even with just 5 bit planes, the proposed method can delineate fundamental structures and primary targets under a fixed low photon flux (PPP = 0.9), albeit with some noise and blurred details. Experiments conducted under extremely low-light conditions (PPP = 0.03) demonstrate the robustness of our method. The proposed techniques will have implications for various passive imaging applications such as consumer photography, microscopy, and astronomy.