.png){kind=icon}
.png){kind=icon}
Return
CDM-CSNet: a conditional diffusion model-based image compressive sensing reconstruction network
.png){kind=icon}
OA
AI
DOI：10.1007/s40747-026-02361-w.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Compressive sensing (CS) has provided a robust framework for reconstructing high-dimensional signals from sub-Nyquist measurements. Recently, the generative power of conditional diffusion models (CDMs) has opened new avenues for solving ill-posed inverse problems. Inspired by these advancements, we propose CDM-CSNet, a novel deep generative framework that integrates the CS measurements as guiding values for diffusion models. Unlike vanilla diffusion models that may produce stochastic hallucinations, CDM-CSNet leverages a dual-condition strategy: it fuses the structural constraints of the CS sampling matrix with the fidelity information of low-dimensional measurements to serve as a non-parametric prior. This integration effectively constrains the generative search space, mitigating the inherent uncertainty of the reverse diffusion process and ensuring that the synthesized outputs are both visually realistic and mathematically consistent with the original signal. Furthermore, the proposed model demonstrates solid scalability, enabling high-fidelity image reconstruction over a wide range of sampling ratios. Extensive experimental results on multiple benchmark datasets validate that CDM-CSNet not only achieves reconstruction accuracy but also preserves intricate structural details that conventional methods often fail to capture.
Keywords:
Compressive sensing
Denoising diffusion probabilistic model
Conditional diffusion model
Deep learning
Residual recovery
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
C
IF:
4.6
Papers:
244
Citations:
0