High Spatial Resolution Digital Imagery
35
Reflectance-calibrated imagery is normally plotted using percent as the unit
of measure for each pixel. Therefore, pixel values would theoretically have a
range of 1 to 100%, assuming that a value of 0% (or DN = 0) would be reserved
for any background pixels. Commercial image processing software generally
does not store and process digital imagery, either reflectance-calibrated and non-
calibrated, using DNs scaled from 1 to 100. The most common digital data for-
mats include:
Binary data where each pixel has a DN of either 1 or 0.
8-bit integer data with a DN range of 256 (28).
16-bit integer data with a DN range of 65,536 (216).
Floating-point data where DNs are stored as real numbers (positive or
negative).
Furthermore, 8-bit and 16-bit data can be stored as signed or unsigned.
Signed DNs include a negative or positive sign with each integer value. Unsigned
DNs include only positive integers. Calibrated image data should be unsigned;
negative DNs should not exist as the theoretical minimum reflectance integer
value is +1. As stated above, the AISA sensor acquires 12-bit data, resulting in a
radiometric range of 4095 (212). The value of 0 is reserved for background pixel.
Therefore, to conform to image processing software file formats, the calibrated
AISA images should be stored in an unsigned (i.e., no negative numbers), 16-bit
format with a minimum reflectance DN of 1 and a maximum DN of 4096. While
this may not appear to be an acceptable scale for percent reflectance, the 1 to
100% scale has simply been re-scaled to 1 to 4095. For example, a reflectance
value of 10% has a 12-bit integer value of 410 (or 409.6 rounded to the nearest
whole number). A 70% reflectance measure is assigned a DN of 2867. Back-
ground pixels should maintain a DN of 0.
The AISA imagery does not follow these assumptions of the theoretical
range associated with 12-bit data. The negative DNs presented a problem. Sev-
eral flightlines (not shown) had reflectance minima in one or more bands ap-
proaching 32,000. A complete assessment of the 38 bands within all flightlines
revealed channels 1, 2, and 38 had the most frequent negative DNs. This radio-
metric problem should be expected from the channels near the extremes of the
range of the sensor CCD. That is, at the upper and lower limits of the CCD, the
detector likely has the greatest signal-to-noise ratios and the lowest relative out-
put (or responsivity). The number of pixels with negative values ranged from
only a few to nearly 10,000. While the larger frequency appears alarming, 10,000
pixels represent less than 0.5% of the total pixels within each image. Finally, the
negative pixels were nearly always assigned to the first and last pixels in the