Multi-layer and pixel-wise guided diffusion model for enhancing single-pixel imaging

doi:10.1016/j.optcom.2025.132409

Multi-layer and pixel-wise guided diffusion model for enhancing single-pixel imaging

2025-09-01

0

PRE

AI

DOI：10.1016/j.optcom.2025.132409

Original

Original request for help

Save

Abstract

En 中文

Improving imaging quality while reducing the sampling rate remains a challenge in single-pixel imaging (SPI). Given that most current research in SPI has not fully utilized the information provided by the measurements, resulting in suboptimal imaging quality, this paper introduces a novel SPI approach based on diffusion model, proposing an algorithm that utilizes measurements as multi-layer and pixel-wise guidance (MPGDM) to enhance SPI performance. Unlike most studies that simply concatenate measurements and noisy images before inputting them into a noise prediction network, we design a specialized noise prediction network called MP-UNet that allows measurements and noisy images to be summarized with the pixel-wise weighted at each layer of the network. Moreover, we introduce a model-driven fine-tuning mechanism, enabling the method to adapt to different physical imaging targets. Simulation and practical experiments demonstrate that our approach outperforms the current state-of-the-art techniques, achieving high-quality imaging across various sampling rates. Additionally, noise impact tests and ablation studies further validate the practicality and effectiveness of the proposed method.

Keywords:

Single-pixel imaging

Diffusion model

Multi-layer guidance

Pixel-wise weighted

Multi-layer and pixel-wise guided diffusion model for enhancing single-pixel imaging

Abstract

Journal

Organization