IEEE Geoscience and Remote Sensing Magazine - September 2014 - 10

the objective of this paper is to provide a tutorial introduction to the most recent image compression standards for
space applications, and give an outlook to future developments aimed at filling feature or functionality gaps that are
not covered by the existing standards.
B. Compression requirements
As has been said, designing a compression algorithm for
on-board use must take into account a number of constraints dictated by the structure of the on-board processing system. Fig. 1 shows the operational mode of an onboard sensor on an airplane.
Such on-board processing system typically has limited
computational capabilities due to power, size and radiation hardening issues. Therefore, low encoder complexity
is highly desirable. In addition to this, the encoder design
must be such that the algorithm should be easy to implement in the available hardware. The typical hardware for
on-board processing has evolved over the years and still
is. The usual preferred choice is a field-programmable gate
array (FPGA), which requires a description of the algorithm in a hardware description language such as VHDL.
Therefore, the algorithm should not employ operations
that are difficult to map to a VHDL description.
Another requirements is the ability to properly handle raw data. What this means is that an on-board compression algorithm will have as input the original digital
numbers generated by the sensor, prior to any processing
other than binning. Therefore, all sensors imperfections,
such as noise, striping, misregistration and so on, which
are typically corrected at the ground segment, will be present in the image. This implies that such imperfections will
yield a loss in compression efficiency, and this loss could
become rather large unless the algorithms are somewhat
robust. E.g., as will be seen in Section II, the CCSDS-123
recommendation defines prediction modes that are robust
towards striping noise.
Moreover, it is known that image transmission from
the remote platform to the ground station can undergo
errors or packet losses. This phenomenon is rather strong
for deep space missions because of the very large distance,

Figure 1. AVIRIS operation mode (courtesy NASA/JPL-Caltech).

and very weak for Earth observation satellites. Nevertheless, since even one packet loss may render the compressed
image file completely undecodable, another requirement
lies in the provision of some kind of error resilience, i.e.,
the image decoding process should not break down completely, nor excessively impair the data, upon occurrence
of occasional errors or packet losses. While there are a lot
of sophisticated error resilience techniques available for
image compression, for Earth observation the most typical
approach is to reset the compression algorithm every once
in a while, so as to create a set of independently decodable
image units, thereby limiting the scope of any information
loss due to communication errors.
It should be noted that the requirements above mostly
apply to on-board compression, whereas another typical
application of compression algorithms is at the ground segment. In this case, however, the requirements are rather different, as there is no significant limitation of computational
power and memory to perform compression, and the objective is also different, namely to distribute the images to the
final users. Since in this case most of the communications
occur via the TCP/IP protocol, which performs retransmissions until the compressed image file has been received
without errors, error resilience is also less important. On the
other hand, some specific issues arise as a consequence of the
way the images are accessed by the users. In particular, users
typically connect to search engines via web browsers or specific software, which allows them to browse the images and
their metadata to facilitate the choice of the products of interest. Given the very large size of these images, this remote
browsing process is only possible if a flexible compression
algorithm is employed, avoiding to send the complete compressed file, since this has a huge size, but sending subunits
of this file that can be employed at the decoder to reconstruct
specific regions of interest that the user has selected dragging
a box in a preview image. Moreover, in order to speed up the
image selection process, it is important that the compression algorithm offers scalability, i.e., it allows sending first a
low-quality version of the subimage of interest, and then one
or more quality improvement layers, so that delays can be
avoided if the user performs an early rejection of a subimage
they have chosen, and moves on to another subimage.
C. struCture of the paper
As has been said, this paper is mostly concerned with
CCSDS standards, which have been designed for onboard compression, while we will briefly discuss the
JPEG2000 standard that is more suitable for compression
at the ground segment. In particular, we will first cover
the two most recent CCSDS standards, namely the new
CCSDS-123 standard for lossless compression of multispectral and hyperspectral images (Section II), and the
CCSDS-122 standard for lossless and lossy compression
of monoband images (Section III). These two standards
cover a lot of possible applications; however, there is still
no standard defined for lossy compression of multispectral
ieee Geoscience and remote sensing magazine

SEPTEMBER 2014

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