orthophoto. Having a redundance of „spectral and
radiometric control points“ the transformation elements
are determined by the method of a least squares
adjustment. Afterwards the PTC-orthoimages can be
determined from all the ClR-orthoimages using the
transformation elements computed in equation (3):
Rerc = a10+ A1 - Ror+ A12- Gorr + A13 - Ber
Gerc= an+ an - Roir+ A2 -Goir+ A23- Boir (4)
Berc= a30 + as - Rer+ as - Goir + A33 - Ber
3.3.3 Classwise determination of transformation ele-
ments: As described in Chapter 2 there is a object spe-
cific reflectance of radiation. A second approach is the
classwise determination of transformation elements:
Rrc,a — d104- a11- Reim, ct 4- A12- Goir, + A13- Ber, a
Grc,« — d209- d21- Rar, a+ az- Ger, a+ az- Ber a (5)
Brc, «1 = A30 + A3 - Rom, 4 + as- Ger, a+ ass- Ber, o
For the described project ten object classes were defined
(water, needle forest, deciduous forest, traffic lines, roofs
light, roofs dark, field with vegetation, meadow, fallow
land, field with sparse vegetation). An experienced inter-
preter visually classified the different classes on the
reference orthoimages by delimiting areas of same
objects. Within each of the ten classes transformation
parameters were determined using class specific
,Spectral and radiometric control points“ and the method
of least squares adjustment.
3.3.4 Land use classification: In order to calculate
PTC-orthoimages, the object information for each pixel is
necessary. For this purpose a maximum-likelihood clas-
sification has been done for all ClR-orthoimages. As
training sets for this computer assisted image classifica-
tion the visual interpreted object areas were used. After-
wards the pixel values of the CIR-colour space were
transformed to the PTC colour space for each object
class separately:
Rerc, d = à10+ A11 - Roir, a + A12- Gerr, 4 + A13- Boir, ct
Grrc, ed = A20+ A21 - Reir, a + a22- Gem, 1 + A23- Beir, 1 (6)
Berc, a = A30 + A31 - Roir, a + az2- Gem, a + A33- Boir, el
3.3.5 Calculation of transformation elements by
means of class centers: The method described in
Chapter 3.3.1 prefers classes with a higher frequency in
the reference orthoimage. This disadvantage can be
avoided by computing the center values of each class in
both colour spaces. Afterwards all the center values are
used as ,spectral and radiometric control points", and the
influence of each class by computing the transformation
elements is weighted equally. Besides this method also
allows an individual ponderation of each class. The
transformation of the CIR values to the PTC values will
be done class independent using equation (4).
3.3.6 Calculation of transformation elements by
means of class centers with subsequent improve-
ment of some classes: The calculation of transforma-
tion elements by means of class centers enables on the
whole good results. Only the colours of few classes are
not transformed correctly into the PTC-colour space T,
overcome this lack by a subsequent improvement 0
single classes can be done. In a first step the transforms.
tion parameters between CIR-colour space and TC-co.
lour space are determined using class centers an
equation (3), and a PTC-image will be calculated, |n 4
second step the quality of the PTC-image will be
controlled and for object classes with insufficient coloy
values the transformation parameters will be calculated
individually using equation (5), and the class specific
pixels are transformed to the PTC-image with equation
(6). The class information of the pixels will be obtained as
described in Chapter 3.3.4.
3.3.7 Simplified methods: Theoretically the transforma-
tion parameters between CIR-colour space and TC-co-
lour space must be calculated for all three bands to get a
PTC-image. Due to the high correlation of the green band
of the CIR-image with the red band of the TC-image
(both corresponding with. the red part of the natural
electromagnetic spectrum) and the high correlation of the
blue band of the CIR-image with the green band of the
TC-image (both corresponding with the green part of the
natural electromagnetic spectrum) the transformation
between the two colour spaces can be simplified:
Brc= ao+ ar- Rer+ az- Ger + as - Ber (7)
and the simulation of the PTC-image can be done by
Rere = Ger
G»rc — Bom (8)
B»rrc — ao-4- ai- Rom d2- Gom 4- a3: Bon
If there was no yellow filter in use to cut off the blue part
of the electromagnetic spectrum during the photo flight
for getting the CIR-photographs, the transformation para
meters can be calculated using the equation below:
Grc — d1o4- d11- Rom + A12 - Gorr + A13 - Ber (9)
Bre = az + a2ı- Rear + a22- Ger + A2 - Bcir
The respective transformation equations read:
Rerc = Ger
Gere = avo + an - Rer+ az - Ger + as - Ber
(10)
Brrc = a2 + az - Rar + A22 - Ger + az3- Ber
Also within this simplified method there is the possibilty
to variate the computation algorithms for the transforme
tion elements such as
* classwise transformation,
* means of class centers and
+ means of class centers with subsequent improvement
of some classes.
3.3.8 Implementation of the algorithms: All described
algorithms in Chapter 3.3.1 to Chapter 3.3.7 were imple
mented on a UNIX computer. For a test image all trans
formations were calulated and investigated. Some of the
472
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
CC m" rr
cm
zm