This research eplored the quality of DIRSIG's simulated images in comparison with real images that were used to obtain truth data. The results of this investigation indicate a great room for improvement of the output that would more accurately represents real data. This calls for a closer examination of DIRSIG's texture application routines, and for alterations to this technique, perhaps even to determine a new technique for the incorporation of texture. Methods for improvment were test, such as expanding the spectral database, and the alteration of DISIG's texture application method to include additional band pass regions.
Careful examination and comparison of
DIRSIG's output images (before new methods were incorporated), to real
image data, revealed a large difference in the amount of information contained
within the scene. The principle components analysis revealed that the
truth or real image contained data over every wavelength, while the DIRSIG
images do not. Also the amount of information continued in the first bandpass
was found to be significantly higher for the real image. Thus there was
determined to be a large gap in the amount of information produced by DIRSIG
for each image.
The expansion of the spectral yielded an increase in the quality of the output image. A qualitative examination of DIRSIG images generated using both limited and expanded databases were completed, in which the expanded database was determined to demonstrate better replication of the characteristics produced by the texture of grass in the HYDICE image. The image generated using a limited spectral database replicated the large structures fairly well but failed to replicate smaller artifacts. A comparison of the DIRSIG image to the HYDICE image demonstrated that DIRSIG has a limited ability to replicate the entire structure of the grass, even with a large spectral database.
There is
an apparent shift in the DIRSIG images toward shorter wavelengths; this may be
an artifact of DIRSIG's sensor configuration for simulation of HYDICE.
Principle components analysis was done on each image to get a quantitative
measure of how much information it contained. The results of the principle
components analysis for the images indicate that there is a great gap in the
amount of information that can be found in the HYDICE image, and the amount of
information that contained in any of the DIRSIG images. The
maximum eigen value for the HYDICE image was 8,649,461. The maximum eigen
values for the DIRSIG images showed a significant difference in the amount of
information present within the images. The HYDICE image also contained data
over every bandpass, while the DIRSIG images each contained a far lower number
of bands containing information.
Using one additional
bandpass appeared to create an increase in the eigen values; incorporating two
additional bandpass regions actually decreased the eigen values. In each case
the increases were quite minor compared to the size of the gap that exists in
the level of information found in the image. Regardless of the number of
bandpasses used, the DIRSIG image still was found to contain a significantly
less amount of data than that contained in the HYDICE image. A qualitative
examination of the resulting images revealed more clearly the improvements that
adding a second bandpass had on the structure of grass in the DIRSIG image.
The
amount of information contained in the DIRSIG images that were examined showed
a continual and significant difference from that of a real image. The methods
for improvement tested through this research did show certain levels of
quantitative improvement, however this improvement was minor. Qualitatively,
when examining the appearance of the DIRSIG images, DIRSIG's ability to
replicate the artifacts found in the real image of grass were significantly
improved with both the expanding of the spectral database and the addition of a
second bandpass region.
DIRSIG's system for
selecting and attempting to reconstruct texture is extremely complex. There is
a large number of bandpasses that may be used to create texture images, and
these may be used in an even larger number of combinations. Determining which
combination would work more effectively is a very difficult task. The factors that
could be used in real life to determine which bandpass regions are numerous,
the most frequent probably being based on which bandpass region are the
scenario would be most concerned with. This research suggests a need for a
closer look at the information that each bandpass regions contains, and how
these bandpass regions interact with one another.
This research, while showing
a great difference in the quality of DIRSIG's images, has pointed a continuous
and large discrepancy in the quantitative replication of texture in the simulated
image. The next step would be to research the effect that the lack of
this significant amount information has within the various systems to be used,
and if there is another viable method for generating the texture that would be compatible
with DIRSIG.