Aerospace and Electronic Systems Magazine June 2017 - 31

Cohenour, Rovito, and van Graas
The difference between the traditional SET model and the revised SET model is the projection is done on demand by the SET
tool instead of being done in advance. This projection process
is the same as in the traditional SET model and just needs to be
moved from a preprocessing stage to the SET.
In the revised SET model, see Figure 1, the raw images are
compressed. The data which now includes the raw images, the
pose, the camera model, and the DTED, are transmitted to the
revised SET. The pose and camera models are extremely small
compared to the images so the size of these parameters should not
impact the transmission or storage of the data.
The DTED must be available to the revised SET. For the
WPAFB 20091021 data set the 40 NITF images were scanned.
The latitude of the top and bottom of the images, and the longitude of the left and right side of the images were recorded. The
latitude span was calculated as the max top less the minimum bottom. Likewise, the longitude span was calculated as the max right
less the minimum left. The latitude range is 0.073° which is 8.1
km. The longitude span is 0.094° which at a latitude of 39.74° is
8.3 km. For a level 2 DTED the spacing is 1 arcsec. The resulting
DTED segment is 339 × 263 points. At two bytes per point the
data is a 206 kB file. This is about 0.5% of a single NITF image.
The DTED data for the image could be included with the raw data,
pose, and camera models, or could be downloaded from generally
accessible Digital Elevation Map repositories. This would make
each frame independent. For a higher resolution elevation map the
DTED data could be included as one image for the entire flight.
The NITF format can be used for the revised model. The six
images can be stored in the NITF file along with meta-data for the
pose, camera models, and DTED. The size of the NITF files would
remain the same as the files for the traditional model.
One other advantage with the revised SET model is that the
camera models, and DTED are retained with the raw images. If a
better DTED, or a better camera model are available when the data
is analyzed, then the geo-registration can be improved. This is not
possible in the traditional SET model since the raw images are not
normally retained.

PROCESSING SPEED
With the traditional SET model, the display of an image is nearly
instantaneous. The SET loads the image, computes the subset of
the image to be displayed, displays the subset, and computes the
latitude and longitude of the displayed subset. The only function
that takes any time is the disk operation to load the image. When
zooming, the image is already in memory and there is no disk access; therefore, the results are even faster.
With the proposed SET model, the image must be reprojected
each time the analyst changes frames, pans, or modifies the zoom.
The projection speed is no faster than it would be for the original projection so there could be substantial delays in generating
the image. This could be unacceptable to the analyst. Faster algorithms or hardware accelerators could be used to reduce the lag
time for the image display. The advantage of the proposed model
is only available at higher levels of zoom, when the display pixel
approaches, or is smaller than the size of the image pixel. In this
JUNE 2017

case, the area to be projected is also small, and can be done without
noticeable delay for the analyst.
The time required to compute the projection is dependent on
the number of DTED points involved, and the size in pixels of the
projected image. For a dense DTED the projection is slower than
for a sparse DTED. For a large area there are more DTED points
that must be projected, and more pixels to project.
An alternate to the proposed Set model is to use the pcolor
projection method to generate a higher resolution image. This image would then be supplied in the NITF format and used with the
standard SET model. The storage size and transmission bandwidth
would go up by the square of the resolution increase. A ten-fold
increase would require 100 times the storage, and data transmission bandwidth. This would eliminate the overhead associated with
doing on demand projection. The time saved doing the projection
would be offset by the increased time required to load the larger
image files.
One way of achieving both low display latency, and high resolution is to use a hybrid system. In the hybrid system, a low-resolution image is projected off line and included with the raw images.
When the set tool is zoomed out the low resolution projected image
is used. When the analyst zooms in the projection is done on demand per the revised SET model.

CONCLUSIONS
The WAPFB data set of 20091021 has been analyzed to determine
the resolution loss due to compression and resampling. For this
analysis, 40 frames were selected from two orbits. The resolution
in the raw image and in the NITF images was measured. Resolution measurement was performed manually using the resolution
target on the abandoned runway at the WFPAB. The line spacing
of the resolution target is not publicly available. The line spacing
was measured using a GE image and OLS.
The loss due to compression was measured two ways. First, the
compression loss was measured by compressing and decompressing the images. The difference between the raw image and the decompressed image was measured directly by subtracting the pixel
intensity values. The results are given in Table 3 and Table 5. The
intensity errors are very small, so we may conclude that any loss
of image quality between the raw image and the NITF image is not
due to compression. Second, the resolution in the reconstructed
image after compression and decompression was measured directly using the resolution chart. Again, the resolution lost due to
compression was minimal, amounting to 5% of the total loss.
The two factors influencing the NITF resolution are compression and resampling. We can therefore conclude that the loss of
resolution is due to resampling.
For a traditional SET model, the images are projected and
stored or transmitted for use by analysts at a later date. Using this
traditional SET model, the resampling error is imprinted on the
NITF image at the start of the process and cannot be recovered
once this has been completed. To resolve this issue, a new SET
model is proposed that performs the reprojection as the image is
being displayed to the analyst. This on demand reprojection can
be performed using the Matlab® pcolor function, or a similar algo-

IEEE A&E SYSTEMS MAGAZINE

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine June 2017