Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Spectral Compressive Imaging via Chromaticity-Intensity Decomposition

Xiaodong Wang, Zijun He, Ping Wang, Lishun Wang, Yanan Hu, Xin Yuan

TL;DR

This work tackles the ill-posed problem of reconstructing hyperspectral images from CASSI measurements under varying illumination. It introduces chromaticity-intensity decomposition, where $\mathbf{X} = \mathbf{C} \odot \mathbf{I}$, to separate lighting-invariant reflectance from smooth illumination, and develops CIDNet, a dual-camera CASSI reconstruction framework that unifies a Hybrid Spatial-Spectral Transformer with a degradation-aware noise estimator. The method includes a theoretically grounded HQS-based optimization/unfolding approach and a Sparse TopK spectral attention mechanism to capture localized spectral structure, along with a degradation-driven per-stage noise model. Experiments on synthetic and real data show state-of-the-art spectral and chromaticity fidelity, highlighting the approach's robustness to lighting changes and its promise for practical, high-fidelity hyperspectral imaging in challenging illumination conditions.

Abstract

In coded aperture snapshot spectral imaging (CASSI), the captured measurement entangles spatial and spectral information, posing a severely ill-posed inverse problem for hyperspectral images (HSIs) reconstruction. Moreover, the captured radiance inherently depends on scene illumination, making it difficult to recover the intrinsic spectral reflectance that remains invariant to lighting conditions. To address these challenges, we propose a chromaticity-intensity decomposition framework, which disentangles an HSI into a spatially smooth intensity map and a spectrally variant chromaticity cube. The chromaticity encodes lighting-invariant reflectance, enriched with high-frequency spatial details and local spectral sparsity. Building on this decomposition, we develop CIDNet, a Chromaticity-Intensity Decomposition unfolding network within a dual-camera CASSI system. CIDNet integrates a hybrid spatial-spectral Transformer tailored to reconstruct fine-grained and sparse spectral chromaticity and a degradation-aware, spatially-adaptive noise estimation module that captures anisotropic noise across iterative stages. Extensive experiments on both synthetic and real-world CASSI datasets demonstrate that our method achieves superior performance in both spectral and chromaticity fidelity. Code and models will be publicly available.

Spectral Compressive Imaging via Chromaticity-Intensity Decomposition

TL;DR

This work tackles the ill-posed problem of reconstructing hyperspectral images from CASSI measurements under varying illumination. It introduces chromaticity-intensity decomposition, where , to separate lighting-invariant reflectance from smooth illumination, and develops CIDNet, a dual-camera CASSI reconstruction framework that unifies a Hybrid Spatial-Spectral Transformer with a degradation-aware noise estimator. The method includes a theoretically grounded HQS-based optimization/unfolding approach and a Sparse TopK spectral attention mechanism to capture localized spectral structure, along with a degradation-driven per-stage noise model. Experiments on synthetic and real data show state-of-the-art spectral and chromaticity fidelity, highlighting the approach's robustness to lighting changes and its promise for practical, high-fidelity hyperspectral imaging in challenging illumination conditions.

Abstract

In coded aperture snapshot spectral imaging (CASSI), the captured measurement entangles spatial and spectral information, posing a severely ill-posed inverse problem for hyperspectral images (HSIs) reconstruction. Moreover, the captured radiance inherently depends on scene illumination, making it difficult to recover the intrinsic spectral reflectance that remains invariant to lighting conditions. To address these challenges, we propose a chromaticity-intensity decomposition framework, which disentangles an HSI into a spatially smooth intensity map and a spectrally variant chromaticity cube. The chromaticity encodes lighting-invariant reflectance, enriched with high-frequency spatial details and local spectral sparsity. Building on this decomposition, we develop CIDNet, a Chromaticity-Intensity Decomposition unfolding network within a dual-camera CASSI system. CIDNet integrates a hybrid spatial-spectral Transformer tailored to reconstruct fine-grained and sparse spectral chromaticity and a degradation-aware, spatially-adaptive noise estimation module that captures anisotropic noise across iterative stages. Extensive experiments on both synthetic and real-world CASSI datasets demonstrate that our method achieves superior performance in both spectral and chromaticity fidelity. Code and models will be publicly available.

Paper Structure

This paper contains 20 sections, 1 theorem, 46 equations, 10 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Degradation Model of CASSI
  4. Optimization Framework of CASSI
  5. Hybrid Spatial-Spectral Transformer
  6. Spectral Attention.
  7. Spatial Attention.
  8. Experiments
  9. Quantitative Results
  10. Ablation Study
  11. Conclusion
  12. Appendix
  13. Multispectral Image Formation
  14. Proof of Illumination-Invariance of Chromaticity
  15. Proof of PAN-Intensity Equivalence
  16. ...and 5 more sections

Key Result

Proposition A.1

In a dual-camera system comprising a CASSI sensor and a grayscale PAN camera exposed under the same illumination $L(\lambda)$, let $s(\lambda)$ denote the spectral response of the camera. The PAN image $\mathbf{I}_{\mathrm{PAN}}(u,v)$ provides a relative estimate of the scene intensity $\mathbf{I}(u where $k$ is a scalar constant that is approximately invariant across spatial coordinates $(u,v)$.

Figures (10)

  • Figure 1: (a) Chrmaticity-Intensity decomposition of HSI images (b) Chromaticity exhibits highlight removal, lowlight enhancement and high-frequency textures.
  • Figure 2: (a) The architecture of our CIDNet with $K$ stages (iterations). (b) The CASSI system uses an intensity-guided mask to modulate the chromaticity. (c) Diagram of asymmetric backbone for our hybrid spatial-spectral Transformer (HSST), with a local window spatial attention (Spa-LWSA) in Encoder and sparse TopK spectral attention module (Spec-TKSA) in Decoder. (d) Details of Spec-TKSA.
  • Figure 3: Simulation HSIs reconstruction comparisons of Scene 7 with 4 (out of 28) spectral channels. The left shows the spectral curves corresponding to the two red boxes of the RGB image. The top-right depicts the enlarged patches corresponding to the yellow boxes in the bottom HSIs. Zoom in for a better view.
  • Figure 4: Simulation: chromaticity reconstruction of Scene 8 with 4 (out of 28) spectral channels. The spectral curves correspond to the two red boxes in the RGB image (top-middle). The top-right depicts the zoomed patches corresponding to the yellow boxes in the bottom chromaticity.
  • Figure 5: Demonstration of sparse and local spectral correlation of chromaticity. Top: RGB contents of the benchmark testing data. Middle: spectral correlation coefficient matrices of the HSIs (28×28). Bottom: Corresponding matrices by the chromaticity.
  • ...and 5 more figures

Theorems & Definitions (2)

  • Proposition A.1: PAN-Intensity Equivalence Under Uniform Illumination
  • proof