IEEE Geoscience and Remote Sensing Magazine - September 2014 - 11

Table 1. aVIRIS and HypeRIon ImageS uSed In THe expeRImenTS. TecHnIcal nameS,
SIzeS and zeRo-oRdeR enTRopIeS (In bpppb) aRe pRoVIded.
SenSoR

name

TecHnIcal name

SIze ^ x # y # z h

enTRopy

AVIRIS

Hawaii Uncalibrated

f011020t01p03r05

614 # 512 # 224

9.09

Maine Uncalibrated

f030828t01p00r05

680 # 512 # 224

8.55

Yellowstone Radiance Sc0

f060925t01p00r12

677 # 512 # 224

10.33

Yellowstone Uncalibrated Sc0

f060925t01p00r12

680 # 512 # 224

12.62

Agricultural Calibrated

EO1H0280342004074110PX

256 # 3129 # 242

10.05

Coral Reef Calibrated

EO1H0830742003120110PW

256 # 3127 # 242

8.46

Urban Calibrated

EO1H0440342002212110PY

256 # 3176 # 242

10.01

Erta Ale Uncalibrated

EO1H1680502010057110KF

256 # 3242 # 242

9.46

Lake Monona Uncalibrated

EO1H0240302009166110PF

256 # 3352 # 242

9.91

Hyperion

and hyperspectral images. In Section IV we take a more
general approach, and discuss a set of possible modifications to the CCSDS-123 standard, including near-lossless,
lossy compression and rate control. Unlike Section II and
III, which aim at describing the main steps of the existing
standards in a tutorial way, Section IV has a more general
scope and reviews more about the underlying theory, highlighting open research issues and possible solutions, as
well as other possible approaches. It has to be mentioned
that, while this paper describes extensions of CCSDS-123,
other options are also being considered, such as a threedimensional extension of CCSDS-122.
D. Dataset
Throughout the paper we will provide compression results
aimed at assessing the performance of the techniques
presented in the paper. To this end, a set of hyperspectral
images will be employed, as shown in Table 1. These are
AVIRIS1,2 and Hyperion2,3 hyperspectral images, publicly
available for download. Technical names and sizes are provided in Table 1, along with the zero-order entropy of the
images. All images are 16 bits per pixel per band (bpppb),
except for Hawaii and Maine that are 12 bpppb. Uncalibrated images are stored as unsigned integers, whereas
calibrated images are stored as signed integers. Considered
calibrated images are radiance images.
For both AVIRIS and Hyperion images, the multi-component volume can be used to produce a spectral signature
for each of the spatial pixels. These spectral signatures are
then employed in remote sensing analysis as classification
or anomaly detection. Fig. 2 illustrates one AVIRIS image
and a spectral signature for some pixels.
II. LOSSLESS DATA COMPRESSION
As pointed out in Section I, we summarize here the most
recent standard for remote sensing data compression. The
CCSDS created the Multispectral Hyperspectral Data
1http://avirir.jpl.nasa.gov
2http://compression.jpl.nasa.gov/hyperspectral/
3http://earthexplorer.usgs.gov

SEPTEMBER 2014

ieee Geoscience and remote sensing magazine

Compression (MHDC) Working Group in June 2007 to
issue a Recommended Standard for Multi- and Hyperspectral image compression. The motivation for creating
this Working Group is explained in this excerpt from [8]:
On-board data compression is needed to make full use of limited spacecraft resources like data storage and downlink capacity.
Multispectral & hyperspectral images can occupy enormous data
volumes, and so compression algorithms specifically designed to
exploit the three-dimensional structure of such images can provide tremendous benefit to space missions.
Thanks to the work of several member space agencies,
observer space agencies and committed participants, the
Working Group was able to deliver CCSDS-123 standard
[9] in May 2012. The standard is intended for on-board
lossless compression of multi- and hyperspectral images.
The coding technique is built upon Fast Lossless (FL) compression algorithm [10] and is able to provide state-of-theart coding performance for a large collection of remote
sensing sensors. As it is oriented towards on-board operation, i.e., in resource-constrained scenarios, the design
was carefully conceived to require a very low computational complexity.

Soil

Water
Vegetation
FIguRE 2. AVIRIS volume and spectral signature at various locations.

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - September 2014