30
ERDC TR-05-1
that little if any radiometric corrections would be required, given that the hyper-
spectral imagery was calibrated, or corrected, to at-sensor-reflectance values.
This suggests that the identical land surface features (e.g., dominant loblolly pine
crowns, soybean fields at the same stage of development) should have identical,
or nearly identical, DNs across all flightlines acquired on the same Julian Date.
If the land surface feature signatures were equal across flightlines after the re-
flectance calibration, adjacent images (e.g., BLINE7E and BLINE8W) would
produce nearly seamless hyperspectral mosaics. This was not the case with the
Blackwater images. Consistently, the seam between two adjacent images was
visually detectable. Radiometric normalization or equalization, involving ma-
nipulation of the image histograms, was employed to radiometrically match
overlapping flightlines.
Histogram analyses generally involves both statistical and graphical tools, as
well some general understanding of the anticipated minima, maxima, and distri-
bution of possible data values (i.e., DNs) within an image-based data set. Nearly
all 38 bands from two overlapping flightlines--BLINE7E and BLINE8W--show
a minimum value of 0, corresponding with the background pixels within each
image file. Raster data, such as digital imagery, must be stored in matrix (i.e.,
rectangular) file formats. Therefore, with non-rectangular output image types,
such as AISA data acquired using a pushbroom scanner system, a rectangle with
dimensions (and accompanying geographic coordinates) slightly larger than the
actual image is created to store the data. The empty, or blank, pixels surrounding
the image data become background pixels, and are generally assigned an output
DN of 0. From an image processing perspective, background pixels add to the
overall complexity of the image statistical characteristics. Figure 20 shows
flightlines BLINE7E and BLINE8W and their associated mean reflectance spec-
tra, including the background pixels. It is good practice to, when possible, re-
move background pixels by trimming the original file using a slightly smaller
rectangle. Some data pixels will likely be lost during the resizing of the file. If
the pixels to be clipped are critical, removing the background pixels may not be
possible. Such was the case with the AISA images. The amount of overlap, more
accurately defined as flightline sidelap, between adjacent flightpaths was too
small to accommodate background pixel clipping because gaps would have been
created. In fact, the percentage of sidelap was so uniformly small that many be-
tween flightpath holidays, or holes, can be seen throughout the completed mosa-
ics. Figure 21 shows flightlines BLINE7E and BLINE8W and their associated
mean reflectance spectra without the background pixels. The means and standard
deviations have shifted considerably.