Abstract <p>Imaging at the single-photon level is a valuable capability in various fields, including microscopy, astronomy, and quantum information science. However, in low-light conditions, quantum shot noise significantly degrades image quality, often requiring extensive averaging over multiple repetitions to obtain usable results. In this work, we explore the feasibility of single-pixel image reconstruction under conditions of extremely low total photon counts, without relying on averaging. Specifically, we employ compressive sampling and examine the impact of shot noise on image reconstruction quality by systematically varying key parameters, including the number of illumination patterns, their spatial structure, and the total number of photons. To ensure uniform pixel sampling, we propose a pattern optimization method based on pixel-weight balancing through the use of heatmaps and permutations. Through computer simulations, we demonstrate that accurate image recovery is achievable even when the mean photon count per image pixel is significantly below one. These results open new opportunities for low-light imaging applications where minimizing light exposure or acquisition time is essential.</p>