Sampling Innovation-Based Adaptive Compressive Sensing

TL;DR

This work addresses high-fidelity image reconstruction under unknown scenes by introducing Sampling Innovation-Based Adaptive Compressive Sensing (SIB-ACS). It defines an innovation criterion $\alpha$ to guide adaptive sampling across multiple stages, forming a negative-feedback ASA framework that reallocates samples toward regions where reconstruction improves most. The framework combines an Innovation-guided Adaptive Sampling Module with a Principal Component Compressed Domain Network (PCCD-Net) that uses proximal gradient steps in both PC and compressed domains to maintain efficiency. Across BSD68 and Urban100, SIB-ACS consistently outperforms state-of-the-art CS methods in PSNR and SSIM, while maintaining manageable computational costs. The approach offers a practical, scalable way to achieve high-quality sensing in challenging, unknown-scene scenarios.

Abstract

Scene-aware Adaptive Compressive Sensing (ACS) has attracted significant interest due to its promising capability for efficient and high-fidelity acquisition of scene images. ACS typically prescribes adaptive sampling allocation (ASA) based on previous samples in the absence of ground truth. However, when confronting unknown scenes, existing ACS methods often lack accurate judgment and robust feedback mechanisms for ASA, thus limiting the high-fidelity sensing of the scene. In this paper, we introduce a Sampling Innovation-Based ACS (SIB-ACS) method that can effectively identify and allocate sampling to challenging image reconstruction areas, culminating in high-fidelity image reconstruction. An innovation criterion is proposed to judge ASA by predicting the decrease in image reconstruction error attributable to sampling increments, thereby directing more samples towards regions where the reconstruction error diminishes significantly. A sampling innovation-guided multi-stage adaptive sampling (AS) framework is proposed, which iteratively refines the ASA through a multi-stage feedback process. For image reconstruction, we propose a Principal Component Compressed Domain Network (PCCD-Net), which efficiently and faithfully reconstructs images under AS scenarios. Extensive experiments demonstrate that the proposed SIB-ACS method significantly outperforms the state-of-the-art methods in terms of image reconstruction fidelity and visual effects. Codes are available at https://github.com/giant-pandada/SIB-ACS_CVPR2025.

Figures (8)

• Figure 1: Comparison of average PSNR performance at the sampling rates of 0.10 and 0.25 on the BSD68 BSD68 and Urban100 Urban100 datasets between the proposed ACS model SIB-ACS, UCS model PCCD-Net and the state-of-the-art CS methods.
• Figure 2: The overview of the proposed sampling innovation-based ASM. (a) Innovation-guided multi-stage AS, (b) Innovation Estimation (IE) based on the reconstructed image information from sampling values before and after Innovation Sampling (IS).
• Figure 3: The overview of the proposed PCCD-Net for image reconstruction. (a) Deep reconstruction process, (b) PCPGD path, (c) CDPGD path, (d) Convolutional block that transitions features from the FD to the CD, (e) Convolutional block that transitions features from the CD back to the FD, (f) Proximal Mapping Module (PMM).
• Figure 4: Visual comparisons of reconstructed image on test003 from BSD68 BSD68 at the sampling ratio of 0.10 and imag048 from Urban100 Urban100 at the sampling ratio of 0.25. The best and second-best results are marked in red and blue colors, respectively.
• Figure 5: Visual comparisons of Uniform Sampling (US) and Adaptive Sampling (AS) on test033 from BSD68 at the sampling ratio of 0.25.