Photon-Efficient Block Compressed Sensing Imaging Using Vision Transformer

Single-photon imaging has become one of the candidates for the next generation of imaging techniques due to its high detection sensitivity. However, the technique faces the problem of mutual constraints between imaging speed and imaging quality, so how to achieve high-quality imaging within a short integration time has become a major challenge. In this work, a photon-efficient block compressed sensing imaging method based on a vision transformer is proposed. By employing the strategy of block sampling and overall reconstruction, high-speed imaging is achieved while preserving high quality. In the experiments, a sampling ratio of 2⁻¹⁰ is achieved with a peak signal-to-noise ratio of 19.46 dB and a structural similarity of 0.762. Moreover, the average photon count per pixel is as low as 0.035 photons/pixel. This work introduces a new strategy for achieving high-speed and high-quality imaging in ultra-low-light environments, offering broad applicability in biomedical imaging, quantum sensing, and other advanced fields.