Saturation in Snapshot Compressive Imaging

2025-01-21

AI Summary

In snapshot compressive imaging (SCI), sensor saturation causes irreversible information loss and severely degrades reconstruction quality. This paper establishes, for the first time, a theoretical characterization of the fundamental performance bound under saturation interference, revealing an intrinsic trade-off between binary mask statistics and reconstruction robustness. We propose a saturation-robust binary encoding mask optimization framework that jointly integrates theoretical modeling, structural constraints on the mask, and a Plug-and-Play (PnP) reconstruction network. Experiments demonstrate that the optimized masks improve average reconstruction PSNR by 2.1 dB under high-intensity illumination, significantly enhancing adaptability to varying saturation levels and mitigating information loss. Our approach provides an interpretable, high-performance hardware–algorithm co-design paradigm for SCI-based video and hyperspectral imaging in strong-light environments.

Technology Category

• Constraint Satisfaction and Optimization: Satisfiability
• Search and Optimization: Combinatorial Optimization
• Computer Vision: Medical and Biological Imaging

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Security and Privacy: Large-scale security measurements
• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization

Abstract

Snapshot Compressive Imaging (SCI) maps three-dimensional (3D) data cubes, such as videos or hyperspectral images, into two-dimensional (2D) measurements via optical modulation, enabling efficient data acquisition and reconstruction. Recent advances have shown the potential of mask optimization to enhance SCI performance, but most studies overlook nonlinear distortions caused by saturation in practical systems. Saturation occurs when high-intensity measurements exceed the sensor's dynamic range, leading to information loss that standard reconstruction algorithms cannot fully recover. This paper addresses the challenge of optimizing binary masks in SCI under saturation. We theoretically characterize the performance of compression-based SCI recovery in the presence of saturation and leverage these insights to optimize masks for such conditions. Our analysis reveals trade-offs between mask statistics and reconstruction quality in saturated systems. Experimental results using a Plug-and-Play (PnP) style network validate the theory, demonstrating improved recovery performance and robustness to saturation with our optimized binary masks.

Problem

• Snapshot Compressive Imaging
• Sensor Saturation
• Image Reconstruction

Innovation

• Snapshot Compressive Imaging (SCI)
• Sensor Saturation Optimization
• Plug-and-Play (PnP) Network