Extreme noise severely degrades image quality in photon-limited imaging systems and challenges existing denoising methods. Classical total variation (TV) models rely on fixed local gradients and often introduce cross-edge smoothing, while deep learning methods may become unstable under extremely low signal-to-noise ratios. To address these limitations, we propose DSGR-Net, a hybrid denoising framework integrating Deformable Sparse-Gradient Regularization (DSGR) preprocessing with neural network restoration. The proposed DSGR model performs adaptive neighborhood regularization by selecting the eight smallest local gradients within a deformable neighborhood for sparse total variation compensation. This structure-aware strategy effectively suppresses noise while avoiding cross-edge diffusion and preserving fine image details. Experimental results on both simulated and real optical imaging data demonstrate that the proposed DSGR-Net achieves improved structural preservation and noise suppression compared with representative denoising methods.

Video Compressive Imaging (VCI) enables low-dimensional detectors to capture high-dimensional data through incoherent encoding. However, traditional pseudo-random coding often exhibits structural sampling that leads to detail loss. While adjusting the sampling rate can balance structured sampling and incoherence, the reconstruction quality remains unsatisfactory. To overcome this limitation, we propose a high-frequency coding method that mitigates the structural problems of pseudo-random coding by reducing low-frequency components. Simulation results show that this method significantly improves image detail reconstruction, with an average peak signal-to-noise ratio (PSNR) increase of 1.6% across various sampling rates. At a 20% sampling rate, the PSNR improvement reaches around 6%. Furthermore, the method integrates easily into existing VCI systems, offering substantial improvements in image reconstruction quality and reliability compared to pseudo-random coding.