🚀 go-pugleaf

In article <simon_wing-010992180153@wings-mac.jhuapl.edu> simon_wing@spacemail.jhuapl.edu (simon ) writes:
>Is there anybody there who has done compression with radar data ?

You must be more specific: "radar data" can mean anything from the raw
signal to SAR range-Doppler images, quad-polarization vector images,
cartographic maps, altimeter profiles, etc. etc. Each can usually be
compressed if you're willing to spend CPU time or sacrifice resolution,
or both.

A raw SAR signal can usually be compressed since its SNR is greatly
increased during subsequent processing (range and azimuth compression).
The Magellan radar uses a 4:1 compression based on a "block floating-
point quantization" algorithm -- data sampled as 8-bit numbers are
split into 8-bit binary exponents and 2-bit mantissae; since the
exponents are expected to vary more slowly than the mantissae, blocks
of data are normalized to an average exponent.  Although this appears
to reduce the dynamic range of the resulting imaged from 256:1 to 4:1,
this is subsequently restored by the averaging that takes place during
range compression and multi-look summation.

4-polarization radar images, i.e. those that combine all senses of
transmit and receive polarization, are usually compressed into 10-
element symmetric Stokes matrix pixels, and then compressed as if they
were simple images.

It is quite difficult to devise an automatic method of compressing
simple images of radar backscatter cross-section. Since they are
usually computer-generated, their pixel dimension and DN range are
likely to be chosen to maximize information content. Secondly, the
dominant noise source is usually "speckle" and is proportional to the
local image intensity. If an 8-bit image appears "sharp" and "gritty"
to the eye, you probably won't get much reduction in size from a 
lossless algorithm such as Huffman or Lempel-Ziv. If you are prepared
to take the loss, run the image through a small median filter and then
use Lempel-Ziv -- but make sure that you aren't throwing away anything
important (e.g. the image of a corner-reflector that you'll need to
calibrate the image intensity!)

This same discussion applies equally to other coherent-source images,
e.g. sonar. Two of the largest imaging datasets currently published on
CD-ROM, the Magellan Venus mapping project and the Gloria side-scan
sonar ocean floor survey, have both elected not to compress their
products.

I'd be very interested to hear from anyone who is working on this
problem.

Regards,
Peter Ford
MIT and Magellan Project