Sample Lossy Image Compression

Examples of Lossy-Compression of 2MASS Atlas Images

R. Cutri - IPAC

23 May 1997

At the current time, it is unlikely that we will have enough spinning disk space to keep on-line the full resolution 2MASS Atlas Images. Therefore, after pipeline processing of 2MASS data, the full-resolution Atlas Images will be archived to tape, and a lossy-compressed version will be kept near- or on-line. The degree of compression that should be employed is still uncertain, and we would like to hear from 2MASS Team members what their opinion is of the appearance and usefulness of images at different compression levels.

The compression algorithm that will be employed is the Hcompress image compression package written by Richard L. White for use at the Space Telescope Science Institute. This software was used to generate the Digitized Sky Survey in its various forms. It is anticipated that the lossy-compressed version of the 2MASS Atlas Images will be the ones accessed by the 2MASS Image Server (see the prototype at ). These compressed images are suitable for qualitative assessment of morphology and positional information about the near-infrared sky. However, they should not be used for quantitative measurement of source and/or background brightness.

We have applied various degrees of compression to a representative sample of Prototype Camera Ks-band Atlas images: the M67 field, an example of a relatively low density stellar field; the core of the Coma cluster of galaxies to illustrate performance on diffuse emission; and the "MSX" (l=30, b=0) field to illustrate the performance in a high source density region. Examples of the original and compressed images of these fields are shown below. It is recommended that you click on the images to view the full-sized JPEG's, or better yet, download the FITS images to get an accurate feeling for the effects of the compression.

A high degree of compression can be tolerated in relatively sparse fields such as M67 and Coma. 20:1 compressed images appear nearly identical to the originals. Even 50:1 compression in the M67 field is remarkably good. The 50:1 compressed version of the galaxy cluster (not shown here) begins to show degradation, although the 20:1 image is quite good. The 20:1 compressed version of the high density MSX field shows considerable information loss for faint sources, and in the background. 10:1 compression compares well with the original, though.

We can adopt a strategy to compress to a nearly constant compression factor, or a variable compression factor that scales with image noise and/or source density. I propose that we use an adaptive compression factor to get the maximum available gains where possible, but to allow the most information in complex areas. One that aims for at least 20:1 compression in sparse fields, and 10:1 in high density ones seems to be reasonable. Please let me know what you think.

Low Density Field

M67 - Original Coadd M67 - 20:1 compressed M67 - 50:1 Compressed
M67 - Original FITS Image M67 - 20:1 Compressed FITS Image M67 - 50:1 Compressed FITS Image

Galaxy Cluster

Coma Core - Original Coadd Coma Core - 20:1 H-compressed
Coma - Original FITS Image Coma - 20:1 Compressed FITS Image

High Density Field

MSX Field - Original Coadd MSX Field - 5:1 Compressed MSX Field - 10:1 H-compressed MSX Field - 20:1 H-compressed
MSX - Original FITS Image MSX - 5:1 Compressed FITS Images MSX - 10:1 Compressed FITS Images MSX - 20:1 Compressed FITS Images