High Spatial Resolution Digital Imagery
41
Regardless of the cause, the illumination anomaly within all of the images
represented a significant problem in creating usable mosaics. The ENVI imaging
processing software offers a cross-track illumination correction. The routine
calculates the overall row (east-to-west) average DNs within an image and then
fits a polynomial to the averages, for each band, to potentially correct (or flatten)
the reflectance data in the cross-track direction (south-to-north). If applied prop-
erly, the darker southern edge of each image should brighten, while the pixels at
the northern edge should darken slightly. This correction routine was applied to
each flightline after removal of the water pixels. The correction algorithm ignores
the background pixels (i.e., DN = 0) when calculating the across-track averages
and when applying the polynomial functions.
3.3.2.5 Histogram Equalization. As a final step in preparing adjacent images
for mosaicking, an attempt was made to equalize the histograms of two overlap-
ping flightlines. This technique uses a simple mathematical formula to match the
general shape of the mean DN curves between two image files. Figure 24
displays the mean and standard deviation DN curves for the 14-band images
BLINE7E and BLINE8W. These curves were calculated using only the overlap
regions of the two adjacent flightlines. The formula employed to match one
histogram to the other is:
⎡⎛ DNin - DNt ⎞
⎤
DNout = ⎢⎜
⎟ σDNr ⎥ + DNr
σDNt
⎣⎝
⎠
⎦
Where
Dnout = output pixel value
DNin = input pixel value
DNt = mean pixel value of transform image
σDNt = standard deviation of transform image
σDNr = standard deviation of reference image
DNr = mean pixel value of reference image.
Flightline BLINE7E was transformed to the reference image BLINE8W.
Within the image processing software, the equation is applied to the 14-band in-
put image (BLINE7E) using matrix algebra. The four mean and standard devia-
tion terms are one-dimensional arrays (or vectors). The output image is an ad-
justed 14-band image. Figure 25 depicts the output mosaic, as both a true-color
and false-color file. While the equalization calculation worked nearly perfectly,