system and are 
IE coordinates 
echnical colour 
hors, modified 
n Broadcasting 
e generated by 
ccording to the 
Q) 
G) 
(4) 
nd considering 
©) 
at A 0- 
(6) 
(7) 
portions). With 
(8) 
cameras, these 
ith the help of 
(9) 
lap, so that the 
ic derivation of 
approaches for 
)IN test colours 
1 showed mean 
[he perceptible 
  
Ralf Reulke 
Using the second optimisation approach brought results for AE of 1.39 and 1.14 for DIN colours and NASA 
reflectances respectively. The solution was validated with an additional 31 NASA reflectances yielding a very good 
result of AE = 0.99. 
The use of the third (non-linear adjustment) optimisation technique yielded even better statistical results, but also more 
extreme outliers and is thus not yet suitable for routine production applications. 
4.3 Differentiation and Data Spread 
The good approximation of the target values on the average still leaves important questions open. On the one hand, it 
should be investigated whether target values that can be differentiated visually can still be differentiated in the camera 
image. Owing to the linearity of the transformation, checking critical colour differences (i.e. the smallest colour 
difference) between target pairs should be sufficient. 
  
  
  
  
  
  
  
  
  
  
  
Target A Target B AE EBU AE 
ADS40 
Yellow Corn Wheat 2.11 3.30 
Flax Birch Leaves 3.93 2.91 
Oats Tomatoes 3.13 2.03 
Unaltered Rocks Quartz Beach Sand 3.23 2.31 
Altocumulus Clouds | Wet Snow 211 2.44 
  
Table 4. Selected results 
The target pairs shown in table 4 are characteristic for these critical cases and show that the magnitude of colour 
spacings is maintained. The critical cases often pertain to colour values of target pairs that, through the perception of 
colour only, cannot be differentiated even with the eye. Overall, it is not expected that the ability to differentiate is 
affected. 
  
|EiReihet 
| æ Reihe2 | 
  
005 115 2 25 3 35 4 45 5 
Delta E 
  
  
Figure 2. Spread of AE 
On the other hand, even for good average AE values, it must be proved that the maximum single deviation remains 
below the visual perception threshold. The histogram analysis of colour deviation between visual perception and camera 
image shows three reflectances with AE = 3.5, three with AE = 4, and one with AE = 4.5 (figure 2, blue). Thus, the 
critical cases lie above the visual perception threshold. This is the case for unchanged filter parameters. However, 
optimisation of the band centres in the spectral channels through adaptive stochastics (within the limits tolerated by 
remote sensing) and simultaneous widening of the top width in the panchromatic channel lead to maximum individual 
deviations of all target values, Emaxs below three (figure 2, red). 
5 CONCLUSIONS 
This paper shows a physical and task related approach for deriving spectral channels for multispectral and true colour 
systems. The multispectral channels are fixed in accordance with the problem to be solved from a scientific standpoint 
and true colour images must be derived from these channels by colour transformation. To improve the appearance of the 
true colour image, the panchromatic channel can be used. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000. 249