Up: Astronomical image compression
From year to year, the quantity of astronomical data increases
at an ever growing rate. In part this is due to very large digitized sky
surveys in the optical and near infrared, which in turn owes its
origin to
the development of digital imaging arrays such as CCDs. The size of
digital
arrays continually increases following the demands of astronomical
research for obtaining larger quantities of data in shorter time
periods. Currently, projects such as the European DENIS and American 2MASS
infrared
sky surveys, or the Franco-Canadian MEGACAM Survey and the American
SLOAN
Digital Sky Survey, will each produce of the order of 10 Tbytes of
image data.
The routine and massive digitization
of photographic plates have been made possible by the advent of
automatic
plate scanning machines (MAMA, APM, COSMOS, SuperCOSMOS, APS, PMM, PDSs)
(Richter 1998).
These machines allow for the quantification of the truly enormous
amount of useful astronomical data represented in a photograph of
the sky, and they have realized the full potential of the
large area photographic sky surveys.
Clearly the storage and manipulation of astronomical data
always requires
the latest innovation in archiving techniques (12'' or
WORM
in the past, CD WORMS or even magnetic disks with RAID technology now,
hopefully DVD in the near future).
In addition, the simple transfer of such amounts of data over
computer networks becomes too cumbersome and in some cases practically
impossible. The
transmission of a high resolution Schmidt plate image
over the Internet would take of the order of 50 hours.
Facing this extraordinary increase in pixel volumes, and taking
into account
the fact that catalogues produced by extraction of information
from the pixels can always
be locally wrong or incomplete, the needs of the astronomer
follow two very different directions:
- On one side the development of Web technology creates a need
for fast access to informative
pixel maps, which are more intuitively understandable than the catalogues
alone.
- On the other side, quantitative work often requires
accurate refinement of astrometry and photometry, or effective redetection
of missed objects.
Thus, the astronomical community is confronted with
a rather desperate need for data compression techniques.
Several techniques have in fact been used, or even developed,
in the field of astronomy.
Véran & Wright (1994) studied
lossless techniques.
White et al. (1992) developed HCOMPRESS, based on the H-transform,
Press et al. (1992) developed FITSPRESS based on the
Daubechies wavelet transform.
JPEG, a general purpose
standard has been tested by us. Compression based on
the multiresolution
pyramidal median transform (PMT) algorithm has been developed by
Starck et al. (1996).
Huang & Bijaoui (1991) introduced MathMorph for
astronomical
image processing.
This issue of data compression
has become more important to us at the Centre de
Données astronomiques de Strasbourg (CDS) in the context of
the ALADIN project, an all-sky "clickable map'' service provided by
Strasbourg Observatory
(Bartlett et al. 1996),
which has been developed with the
goal of storing and
providing network access to a full sky archive of digitized Schmidt
plates.
Currently we provide access to this archive on the Web,
allowing
users of the CDS services to retrieve images of objects starting from
occurence of names in astronomical catalogues, the SIMBAD database,
and bibliographical references such as
titles or abstracts.
In previous papers
(Starck et al. 1996;
Murtagh et al. 1998)
we reported our findings concerning
the effects of three compression algorithms on astrometry and photometry.
The three methods considered were HCOMPRESS, FITSPRESS and the video
standard JPEG. Concerning the signal-to-noise ratio, the photometry and
astrometry, JPEG and HCOMPRESS produce images of equivalent quality, but
FITSPRESS is worse than the other two methods. The conclusion of this study
is that the standard JPEG method was, ultimately, not so bad, even if block
artifacts appear. However such block artifacts certainly appeared to be
prohibitive for compression rates above 40.
In the present work, we extend our research to more realistic
astronomical settings by considering stars with pixel values closer to
the background, and by examining different criteria for
position and magnitude determination. In addition to the three
previous compression
methods we studied two other methods: PMT, based on multiresolution
analysis,
and another one based on mathematical morphology, both implemented in
the MR/1
package (MR/1 1998). We used a sample of nearly 2000 stars
from an ESO Schmidt
plate (not one, in fact, belonging to the survey plates)
centred on the globular
cluster M 5. The results indicate that PMT can give compression ratios
of up to 5 times the maximum ratio obtained from the other methods, when the
regions are not too dense.
The next section contains
a brief and general description of image compression techniques, and
of the four compression software packages, FITSPRESS, HCOMPRESS, JPEG and PMT.
This is followed in Sect. 3 by a presentation of the data and
calibrations
used for our study (and a discussion of our approach to testing
the astronomical quality assesments of the compressed images), and a
presentation
of our results. We then conclude in Sect. 4.
Up: Astronomical image compression
Copyright The European Southern Observatory (ESO)