IEEE Geoscience and Remote Sensing Magazine - September 2014 - 16

SNR Energy (dB)

55
50
45
40
35
30
25
20
15

JPEG2000
CCSDS 122.0
0

0.5

1.5
2
2.5
Rate (bpppb)

3.5

Figure 8. AVIRIS Yellowstone uncalibrated Sc0 band 99. Rate-

distortion comparison.

SNR Energy (dB)

two that depends on the dynamic range (magnitude) of
all wavelet coefficients in that segment (including AC
coefficients). Then the difference between consecutive quantized DC coefficients (in raster scan order) is
computed. This difference is mapped to a non-negative
integer. Every such 16 consecutive integers are marked
as belonging to the same gaggle, and are then jointly
entropy-coded. These mapping and entropy-coding are
based on the CCSDS Lossless Compression standard
[12]. For entropy-coding, there are two choices, either
a fixed-length code or a variable-length code; the former is faster, while the latter yields improved coding
performance.
2) After sending the first bits of the quantized DC coefficients, the non-quantized DC coefficients might still
have some bit planes above the number of bit planes
needed by the largest AC coefficient in that block.
Some of these remaining bit planes are then sent on a
bit plane by bit plane fashion.
3) Next, the bit depth of the AC coefficients in each
block is sent. In fact, similar to what happens for the
DC quantized coefficients, the difference between the
highest bit plane depth-of all AC coefficients-in
two consecutive blocks within a segment is computed,
then mapped to a non-negative integer, and then jointly entropy-coded for all gaggles.
4) Last, if there are any, the remaining DC coefficients bit
planes along with the bit planes of AC coefficients are
sent also on a bit plane by bit plane basis using several
refinement passes.

55
50
45
40
35
30
25
20
15

JPEG2000
CCSDS 122.0
0

0.5

1.5
2
2.5
Rate (bpppb)

3.5

Figure 9. Hyperion Lake Monona uncalibrated band 48. Rate-

distortion comparison.
16

Within a segment, the bitstreams of the different blocks
are interleaved, so that the overall quality of the recovered image is improved (it is expected that sending some
information of all spatial areas of the image is better than
specializing in a reduced spatial area). Notice also that
the bitstreams for each 8 # 8 block are associated with the
other blocks in the same gaggle, and that it is not possible
to decode a single block. In addition, because of the mapping to non-negative integer values and entropy-coding
employed for both DC coefficients and AC coefficients
highest bit depth, a gaggle can not be independently
decoded either, and information from the whole segment
must be retrieved.
B. MonoBand ExpEriMEnts
We now here report the progressive lossy-to-lossless coding performance of CCSDS-IDC for the images in the
considered data set. Figs. 8 and 9 provide a rate-distortion comparison between two coding standards, classical JPEG2000 [29] and CCSDS-IDC, for one component
of AVIRIS Yellowstone (band 99) and Hyperion Lake
Monona (band 48).
None of these two coding techniques apply any multicomponent transform along the spectral dimension. As for the
spatial wavelet transform, JPEG2000 applies 5 levels while
CCSDS-IDC is restricted to 3 levels. Both use a float 9/7 discrete wavelet transform. The figures plot the bitrate (in bits
per pixel per band) versus the Signal to Noise Ratio (SNR),
measured in dB, considering the original energy of the image.
Kakadu software [21] has been used for JPEG2000 encodings. Delta software [30] has been used for CCSDS-IDC.
The reported results indicate that at very low bitrates
(large compression ratio), JPEG2000 benefits from the use
of a 5-level DWT. However, from 0.25 bpppb onwards, the
performance is rather close. The results for the other components of these two images as well as for the other images
in the data set are consistent with these findings.
As happened with CCSDS-123 standard reviewed in previous Section II, the CCSDS-IDC recommended standard is
able to provide a highly competitive rate-distortion performance for remote sensing data when applied on-board as
compared to more complex coding techniques.
Fig. 10 shows component 99 of the original AVIRIS Yellowstone uncalibrated image (Scene 0), as well as the recovered image when it has been encoded at 0.1 and 1.0 bpppb.
Fig. 11 shows a zoomed-in area of the same image at the
same bitrates. A bitrate of only 0.1 bpppb seems scarce for
most remote sensing applications, but the quality of the
recovered image at 1.0 bpppb, while still presenting visible
artifacts, is notably higher.
C. Multi-CoMponEnt CoMprEssion
In the case of lossy compression of multi-component
images, there is no available CCSDS recommendation at the time of this writing, although the CCSDS is
working towards the definition of a lossless and lossy
ieee Geoscience and remote sensing magazine

SEPTEMBER 2014

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