Single disperser design for compressive, single-snapshot spectral imaging

Ashwin A. Wagadarikar, Renu John, Rebecca Willett, David J. BradyProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 2007 · DOI

TL;DR

This paper introduces a novel "single disperser" spectral imaging system that, leveraging compressed sensing principles, captures a full 3D spectral datacube in a single snapshot using fewer optical components than previous designs. This simplified, static system promises more efficient and accessible spectral imaging for various applications.

What Problem Does This Paper Address?

Traditional spectral imaging systems capture a 3D data cube (2 spatial dimensions + 1 spectral dimension), but doing so efficiently and in a single snapshot has been a challenge. Existing single-snapshot methods like Computed Tomography Imaging Spectrometers (CTIS) and Dual Disperser Imaging Spectrometers (DDIS) often involve complex optics, intricate alignment, or limitations in capturing certain scene features (e.g., DDIS struggles with point sources). This paper addresses the need for a simpler, more robust, and versatile single-snapshot spectral imager that can reconstruct the full spectral datacube by exploiting the inherent sparsity of natural scenes in a transformed domain, a concept enabled by compressed sensing.

Methods

The core innovation is the "Single Disperser Imaging Spectrometer" (SDIS). This system employs a novel static design featuring:

  • Coded Input Aperture: This aperture modulates the incoming light based on both spatial and spectral information.
  • Single Dispersive Element: A prism or similar element spreads the light spectrally.
  • Detector Array: Captures the combined spatial and spectral information.

Unlike previous methods that might separate spatial and spectral information more distinctly, the SDIS intentionally mixes them on the detector. This mixing, guided by the coded aperture, creates "coded projections" of the spectral datacube. The system then relies on a compressed sensing reconstruction algorithm (specifically, the Gradient Projection for Sparse Reconstruction - GPSR method) to demix these projections and reconstruct the full spectral datacube. This approach requires fewer optical elements and simpler alignment compared to dual disperser designs.

Datasets

  • Simulated Dataset: A 256x256 RGB image of peppers was converted into a synthesized 15-channel spectral datacube to test the system's reconstruction capabilities.
  • Experimental Dataset: A ping pong ball illuminated with two lasers (543 nm green and 632 nm red) was used to generate a measured detector output for experimental verification.

Key Insights and Ideas

  • Simplicity through Compression: The SDIS demonstrates that by intelligently mixing spatial and spectral information at the detector and employing compressed sensing, a simpler optical system can achieve the same goal as more complex ones.
  • Sparsity is Key: The reconstruction algorithm's success hinges on the assumption that spectral datacubes have a sparse representation in a wavelet basis. This means most of the information can be represented by a few significant coefficients, allowing for reconstruction from undersampled measurements.
  • Trade-offs in Design: While the SDIS is simpler, the mixing of spatial and spectral information means a raw detector measurement doesn't directly reveal spatial structure, unlike some other methods. This necessitates the reconstruction step.
  • Potential for Point Sources: The paper highlights that the SDIS design, unlike the DDIS, can reconstruct the spectrum of a point source, expanding its applicability.

Key Results

  • Simulation: The simulation successfully reconstructed a spectral datacube from a synthesized scene, producing an RGB image with some artifacts but demonstrating significant potential.
  • Experiment: The prototype SDIS was able to reconstruct spectral content from a ping pong ball illuminated with two lasers. The reconstruction of spectral features over a range of 535 nm to 640 nm took approximately 18 minutes.
  • Optical Distortion Impact: Experimental results showed that optical distortions in the system, particularly at the edges of the field of view, negatively impact reconstruction accuracy.

Limitations and Future Work

  • Unoptimized Code Pattern: The code pattern used in the experiments was not optimized, suggesting that finding a pattern with better auto-correlation properties could improve reconstruction.
  • Optical Distortions: The current optical system exhibits distortions that blur mask features and hinder accurate reconstruction. Future work will focus on developing an optical system that minimizes these distortions and increases the field of view.
  • Calibration: The experimental results showed a wavelength-dependent stretch that the reconstruction algorithm didn't fully account for. Future work suggests the need for a calibration datacube that captures the aperture code pattern across all wavelengths for more accurate reconstructions.
  • Realistic Scenes: The authors aim to experimentally verify the SDIS's ability to generate accurate datacubes of more complex, realistic scenes with piecewise smooth spatial and spectral features.

Summary updated

Abstract

Recent theoretical work in "compressed sensing" can be exploited to guide the design of accurate, single-snapshot, static, high-throughput spectral imaging systems. A spectral imager provides a three-dimensional data cube in which the spatial information of the image is complemented by spectral information about each spatial location. In this paper, compressive, single-snapshot spectral imaging is accomplished via a novel static design consisting of a coded input aperture, a single dispersive element and a detector. The proposed "single disperser" design described here mixes spatial and spectral information on the detector by measuring coded projections of the spectral datacube that are induced by the coded input aperture. The single disperser uses fewer optical elements and requires simpler optical alignment than our dual disperser design. We discuss the prototype instrument, the reconstruction algorithm used to generate accurate estimates of the spectral datacubes, and associated experimental results.
BibTeX
@inproceedings{wagadarikar2007single,
  author = {Ashwin A. Wagadarikar and Renu John and Rebecca Willett and David J. Brady},
  title = {{Single disperser design for compressive, single-snapshot spectral imaging}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {2007},
  doi = {10.1117/12.734058},
}
Figure