Multiscale Reconstruction of Photon-Limited Hyperspectral Data

  • Authors:

  • Kalyani Krishnamurthy;Rebecca Willett

  • Affiliations:

  • Department of Electrical and Computer Engineering, Duke University, Durham, NC 27708;Department of Electrical and Computer Engineering, Duke University, Durham, NC 27708

  • Venue:
  • SSP '07 Proceedings of the 2007 IEEE/SP 14th Workshop on Statistical Signal Processing
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper combines recent innovations in intensity estimation for marked Poisson processes and multiscale nonparametric function estimation using generalized linear models (GLM) to perform photon-limited hyperspectral image reconstruction. Inspired by spectroscopic images of solar flares, which are very intense at some energies but very weak at others, we develop a multiscale intensity estimation method which can adapt to spatial inhomogeneities, spectral emission lines, and large spectral dynamic ranges. This approach exploits the fact that boundaries between different physical structures exist at every spectral band, even when the contrast is very small in some of those bands. Incorporating this denoising method into a generalized expectation-maximization method allows very faint features to be accurately reconstructed from blurry, photon-limited observations.