Dual-branch wavelet diffusion for single-pixel imaging

Single-pixel imaging is a technique that utilizes spatial light modulation and a single-point detector to reconstruct images through computational methods. However, under ultra-low measurement rates, existing reconstruction methods still face significant challenges, such as the loss of high-frequency details and noise interference. This paper proposes a wavelet-domain dual-path reconstruction method named WDNet. This approach decomposes the image into high-frequency and low-frequency components in the wavelet domain and reconstructs them separately: the high-frequency components are restored through a lightweight network to recover texture details, while the low-frequency components guide a diffusion process to maintain structural stability. The advantage of this design lies in its ability to address the characteristics of different frequency components in a targeted manner. Experiments demonstrate that under extremely low measurement rates, the proposed method achieves clearer and less noisy reconstruction results in both simulated data and real acquisition systems.