230 
3. EXPERIMENTS 
In this section, results of the experimental investigation into area based colour matching are presented. It is the 
first test for the multichannel algorithm proposed above. Of main interest are the questions on the quality of 
the simultaneous colour matching compared to results of the single channels and, as an alternative, to results 
applying colour transformations and taking the intensity or brightness channels for matching. 
A discussion of colour representation and colour transformation to different colour spaces is beyond the scope 
of this paper. From scanning we get RGB images which can be transformed to HSI and YIQ. HSI refers to hue, 
saturation and intensity which are the characteristics generally used to distinguish one colour from another. 
Hue is an attribute which represents dominant wavelength as perceived by an observer. Saturation refers to 
relative purity i.e. the amount of white light mixed with a pure colour. Brightness or intensity are most likely 
comparable with the grey values of a panchromatic image. The conversion from RGB to HSI is quite simple 
for the intensity component, just defined by I = (R +G +B)/3. The relations for the saturation and hue are 
somewhat more difficult. The YIQ colour model is used in colour TV broadcasting. One purpose for YIQ is 
to maintain compatibility with monochrome TV standards. The luminance Y, which is proportional to the 
amount of light perceived by the eye, and the colour components (I, Q) are decoupled. The colour transforma- 
tion from RGB to YIQ is defined by a matrix multiplication. Restricting to the Y component the conversion 
is: Y= 0.229 R + 0.587 G + 0.114 B. Obviously, I and Y are determined by weighted averaging of the RGB 
components with different weights for Y and equal weights for I. For more details please refer to textbooks on 
image processing. 
3.1 The material used for the experiment 
For the experiments an RGB image pair and an infrared image pair have been used. À summary is given in 
table 1. Both image pairs are colour diapositives. The film type of the infrared images is a false colour type. 
'The spectral bands IR, red, green are mapped to the colour components red, green, blue of the false colour film. 
  
  
  
  
photo flight photo scale focal length film material scanned with scan mode 
Glandorf 1:8 000 153 mm RGB colour 30 um density 
Aug. 1991 diapositive (y-correction) 
Schorndorf 1:6 000 305 mm colour IR 30 um transmissivity 
July 1991 diapositive (linear mode) 
  
Table 1: Summary of the material used. 
The images are scanned with the Zeiss-Intergraph scanner PS1, in which colour images are digitized in three 
separate passes using appropriate RGB filters. Prerequisite for scanning is a radiometric calibration of the CCD 
with respect to a dark and bright calibration patch (also called normalization) and a basic colour filter calibra- 
tion. In the latter case a relative colour response table is adjusted which corrects for different transmissivity of 
the colour filters. 
A transfer function for converting the internally captured 13 bit CCD input data to 8 bit output image data 
allows for an individual radiometric adaptation to an image. In practice one has to adjust to dark and bright 
local regions of an image which relates to minimal and maximal transmissivity parameters. The 13 bit to 8 
bit transformation takes these parameters into account. The transformation function is linear if the image is 
scanned in the transmissivity mode and logarithmic if the density mode is chosen for scanning. The logarithmic 
curve increases contrast in the dark areas and reduces contrast in the bright regions. 
It is not easy to clearly define a procedure for scanning images with excellent radiometric quality. À main 
reason is that the examination on the quality mainly relies on the subjective visual impression of the operator. 
In practice, support is merely given by the histograms of the three colour components. Furthermore for different 
tasks the scan parameters might be chosen differently. Because scanning influences the colour image quality an 
impact on the quality of matching colour images has to be expected. In this respect the investigation on colour 
matching might be extended to an investigation of scanner properties. The theoretical possibility of scanning ' 
colour images by a colour specific adjustment of the transmissivity parameters is not taken into account. For 
each project the three colour channels of both images were scanned with the same parameters. 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995