Reconstruction Applied to
Inclusive Dates: 04/16/02 - 08/16/02
Background - A need for astronomical observations to push the limits of resolution and detection sensitivity always exists. Although one method is to build larger telescopes and new instrumentation, by far the most efficient, immediate, and cost-effective approach is to optimize the data processing to make the best use of images obtained with the current telescopes and instruments. The main limitation in ground-based observations is the seeing of the atmosphere, that is, the blurring of the images by the turbulence of the Earth's atmosphere above the telescope. Even space-based telescopes suffer from blurring effects due to limitations of the optics, errors in tracking, and other factors. All these effects contribute to what is called the point-spread function (PSF), which is what an idealized point source target (for example, a distant star) looks like in the final image. The PSF is a measure of how well the instrument focuses all of a source's photons into a very small area, and is the limiting factor in the instrument's ability to distinguish two closely-spaced objects or the surface of a resolved object (such as the surface of an asteroid), or the detection of faint sources above the background noise.
Post-processing reconstruction methods attempt to improve the PSF, and therefore improve resolution and limiting magnitudes. The basic premise behind most image reconstruction techniques is that the ``blur'' introduced into an image by the Earth's atmosphere and optics of the imaging system is equivalent to a convolution of the true image with some PSF. The purpose of this quick-look internal research project was to make a proof-of-concept demonstration of using the novel image processing and reconstruction technique "PIXON" to solve a variety of these astronomical image-processing problems. This project is to enable SwRI to develop resources and unique capabilities in observational solar system, stellar, and extragalactic studies.
Approach - The team obtained a ground-based image of the 30 Doradus star cluster (also known as the Tarantula Nebula) in the Large Magellanic Cloud (a neighbor galaxy to the Milky Way). After processing the image with PIXON, the team compared the results with a Hubble Space Telescope (HST) image of the same field.
This image was selected for several reasons, all to make it a difficult (but in many ways, typical) dataset that PIXON might have to handle in general use. First, is the extreme crowding of stars in the image; most of the washed-out white blobs are, in fact, clusters of closely packed stars. Second is the high level of nebulosity, which contributes to two difficulties in image processing: a variable background that makes it hard to determine a background sky value, and points of bright and compact nebulosity that can sometimes be mistaken for stars (several of the white, irregularly shaped blobs are such nebular regions as opposed to stars). Third, the image has several detector artifacts, for example, bad columns or individual pixels that may be bright or dark. Such defects are common among the digital detectors (charged couple devices; CCDs) used in astronomical observing, so any image-processing method needs to be able to robustly deal with these artifacts.
Accomplishments - The agreement is remarkable, especially considering the fact that the PIXON image is from a 3-second exposure obtained with a 0.9-meter ground-based telescope, and the HST image is a 23-second exposure obtained with a 2.4-meter telescope in space! The PIXON image has resolution comparable to HST, and recovers surprisingly many of the stars that appear on the HST image, including stars buried in highly variable nebulosity. Note that although the HST image appears to have more stars (and it does, particularly because of the nearly eight times longer exposure and the more than seven times greater collecting area), many of the small points in the HST image are actually not stars, but cosmic ray events (a common CCD artifact). In addition, there are notably few false detections in the PIXON image, showing that it is reasonably robust.