|
Raw data courtesy of
NRL/SOHO/LASCO |
CESDIS/GSFC |
Figure 1: LASCO-CI Preliminary Results
The research objective is to design, develop, and apply image restoration, image deconvolution, spectral restoration, and hyperspectral restoration methods to space flight data on massively parallel computers.
The model-based approach adopted here seeks to combine, in an optimal fashion, all prior knowledge related to the problem under study. Prior knowledge refers to any and all known information about the optical response, detector response, imaging process, noise statistics, and class of object being viewed. The approach is model-based since it requires a model of the sensor optics, detection system, and noise model. Model based image processing generally involves iterative non-linear algorithms and is computationally expensive but lends itself well to massively parallel computing due to the inherent parallelism
A number of model-based maximum entropy methods have been researched, developed, and successfully applied to space flight data. The methods are all primarily based upon maximum entropy techniques. These algorithms attempt to utilize all the prior knowledge related to the system and the sensor detection process to extract the maximum scientific content from the observed dataset. The techniques have been successfully applied to: (i) restore aberrated Hubble Space Telescope - Faint Object Camera images, and to (ii) separate line emission images from continuum images for the Solar and Heliospheric Observatory (SOHO) Large Angle Spectrographic Coronagraph (LASCO).
The SOHO/LASCO example is shown in the figures. The LASCO instrument is a coronagraph that only views the solar corona by occulting the solar disk and removing the diffracted and scattered radiation by a series of occulting masks, pupil stops, and baffling structures within the telescope. Furthermore, LASCO has an internal Fabry-Perot etalon to allow very fine wavelength tuning. Even with the coronagraph the images still suffer from a very bright background due to the diffracted and scattered light. Thus the fine structure to be viewed is embedded in a very bright background.
In order to enhance the fine structure the etalon is tuned to observe four images at a solar spectral line (line + continuum emission) and four images observed with the etalon shifted off the spectral line (continuum emission). The line emission images contain both the stray light (scattered and diffracted) and the coronal features of interest and are dominated by the stray light. The continuum emission images contain only the stray light. Ideally one would like to separate the line emission from the continuum emission. One approach is to average the four line emission images and difference them from the average of the four continuum emission images. The left image of Figure 1 shows an example of this differencing technique. The right image of Figure 1 shows the result of applying model-based maximum entropy with maximum likelihood constraints to the set of eight observed images to simultaneously sharpen and separate the line from the continuum emission.
The algorithm is model-based in that it relies on an optical wavefront propagation model of the LASCO instrument. The optical response functions (PSF's), from the wavefront propagation model, are shown in Figure 2, at 1, 2 and 3 solar radii. "EE" refers to the ensquared energy and represents the maximal fractional amount of energy that can fall into a single pixel. Note the significantly enhanced structure within the solar corona.
Figure 2: LASCO-CI Calculated Optical PSF's
This work is significant in that it shows that with proper understanding of the sensor system and the correct algorithms and computational resources one can significantly enhance the imaging quality of an optical sensor.
We are beginning to assess the possibility of applying these techniques to the SOHO/Coronal Diagnostic Spectrometer and also to the next generation of GOES weather satellites.
Richard G. Lyon
Center of Excellence in Space Data and Information Sciences (CESDIS)
and
Department of Electrical Engineering, University of Maryland
lyon@nibbles.gsfc.nasa.gov
301-286-4302
URL:
http://cesdis1.gsfc.nasa.gov/~lyon/
John Dorband
Goddard Space Flight Center
dorband@nibbles.gsfc.nasa.gov
J. Michael Hollis
Goddard Space Flight Center
hollis@achamp.gsfc.nasa.gov
Table of Contents | Section Contents -- Applications | Subsection Contents -- Inhouse Computational Scientists
