Stefan Diener
with
res : range of gray values
bri ght : gray value of bright image at position rc (row, column)
dark, : gray value of dark image at position rc
gray, : gray value at position rc of the original image data, which has to be corrected
original
grav, : gray value at position rc of the corrected image data
corrected
Applying the respective gain and offset to each pixel leads to the corrected gray value:
— gain, * gray, —offset,. (2.2)
original
gray,
corrected
With one set of gain and offset values there is the possibility to correct images of one state of the system. This means if
one of the above mentioned influences changes the set of gain and offset values changes too. Therefore one has to store
a set of gain and offset values for each state of the system. This means for every temperature step (e.g. in steps of 5
degrees from —20°C to + 20°C), each used aperture (e.g. 4, 5.6, 8, 11) and each used TDI (e.g. 0 to 7) and especially
each combination of these states one has to store the gain and offset values.
Using a simple method to store one look up table (LUT) for gain and offset respectively we get the following needed
storage capacity assuming that gain and offset are stored with 2 bytes. Using e.g. a CCD array with 4k by 4k pixels the
following space would be needed:
4096 * 4096 * (2 Bytes) *9 * 5 * 8 = 11.25 GByte
One can easily see that using the classical LUT method the needed storage for the normalisation data explodes. Thus we
need an alternative method.
2.2 DMC Normalisation Approach
Since the DMC camera has not only one CCD array this topic becomes more critical. With increasing number of
cameras the amount of normalisation data increases too. Each CCD has an own behaviour with regard to the sensitivity
of each pixel. Using LUTs there is no way to store the influences.
Thus we develop another strategy for the radiometric normalisation. The main purpose of this strategy is to use only one
LUT for each camera. That way one can inhibit the strong increase of the needed storage. But this results in a new
problem. We work out the coherence between the changes of the CCD values and the current examined influence. At
the end we get a function which delivers the correct gain and offset value for a given input state. For each of the shown
influences one function is computed.
Instead of examining the gain and offset values it is also possible to look at the changes of the dark and bright image.
There is no difference in the needed amount of storage if the gain and offset values are stored instead of the dark and
bright image. From the latter one can compute gain and offset. The reason for this is situated in the fact that more easily
a functional model can be found if the modifications in the light or dark image are taken.
Then if these functions are determined, one has simply to transform the original gray value at certain position in the
dark and bright image with these functions sequentially to get the correct gray values of dark and bright image. With
these transformed dark and bright image one can compute the gain and offset of the current state. With equation 2.2 the
correct gray value of a given image can be computed afterwards.
3 COLOUR COMPOSITE GENERATION
3.1 Workflow
The generation of the colour composite can be divided into five steps as shown in figure 2. At first the offsets of the
multi-spectral cameras have to be corrected. These offsets were estimated during the calibration of the DMC and result
in parameters for affine transformations. Furthermore the individual multi-spectral channels have to match the mosaic
image, so another affine transformation is necessary. The post-processing software will combine the parameters of both
transformations to perform only one affine transformation (and interpolation).
84 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000.
—
After
resoli
To pi
the Ir
To mr
imag
done
biline
The |!
inter]
more
imag
Aftei
infor
desci
Now
moss
the d
32
We c