ables on the
/ Classes are
JUI Space.To
rovement of
> transforma.
| and TC-co.
centers and
culated. In à
age Will be
ficient colour
9e calculated
lass specific
vith equation
> obtained as
> transforma-
and TC-co-
ands to geta
e green band
1e TC-image
the natural
elation of the
band of the
in. part of the
ansformation
fied:
(7)
done by
(8)
the blue par
> photo flight
mation para
below:
(9)
(10)
he possibilty
| transforma
improvement
All described
were imple
ige all trans
Some of the
|
|
I
|
results (three PTC-images) as well as the original CIR-
input image, the TC-(reference)input image and the class
map of the classification are illustrated in colour within
these proceedings.
4. RESULTS
The quality control of the results of the CIR- to PTC-
image transformation and with this a judgement of the
chosen algorithm was a weak point within the presented
project: The colour infrared photographs and the
corresponding reference true colour photographs were
not taken in the same year, not in the same season, not
at the same time and not with the same camera ob-
jective. Due to this fact there are some objects in the
photographs which are not identical (e.g. there are some
fields that have different vegetation status). This
circumstance was considered by delimiting the object
classes to determine the transformation parameters, but
it was not taken in consideration by the computation of
statistical values as presented in Chapter 4.2. So the
results are made worse.
The quality of the PTC-image depends on following
influences:
= quality of photographs
= contents of photographs
= amount of distinguished classes
= selection of training areas for the classification
= kind of transformation
In the presented paper two possibilities of quality control
are outlined. For one test area CIR-images were trans-
formed to PTC-images and afterwards compared with
theTC-image of the same area. When judging the results
it is to take into account that in this case the CIR and the
TC photographs were not taken at the same time, as
noted above.
4.1 Visual quality control:
Every visual quality control is subjective: Each person
has a different preference of contrast, brightness and
colour balance of a photo and so within this paper only
the impressions of the authors are outlined in table f:
class class center | classwise
center | with classwise | (method
(method | improvement 3.3.3)
3.3.5) (method 3.3.6)
transformation
of one band e e
- (equation 9)
transformation
of two bands e e
- equation 7) *
transformation
of all bands e e
equation 3 +
table 1: visual quality control
Legend of table 1: - bad, e neutral, + good
473
4.2 Quality control with statistical values:
For the quantitative quality control the corresponding
pixel values of the colour bands of the PTC-image
(orthoimage)
and of the
original
TC-image were
subtracted. The mean value of each colour band and the
corresponding variance are outlined in table 2 to 4. As
exspected, due to the high degree of freedom, the
classwise transformation of CIR-pixel values to PTC-pixel
values gives the best results.
class center | class center classwise
method with classwise method
335 improvement 333
method 3.3.6
Red band -7.4/54.6 -7.4/54.6 -7.4/54.6
Green band | 30.2 / 48.4 30.2 / 48.4 30.2 / 48.4
Blue band | 10.1 /38.0 10.0 / 37.9 -3.1/34.0
table 2: quality control with statistical values - transfor-
mation of one (blue) band (equation 9)
(mean value / variance of differences between TC and
PTC pixel values)
class cen- | class center | classwise
ter with classwise | method
method | improvement 3.3.8
335 method 3.3.6
Red band -7.4/54.6 -7.4/54.6 |-7.4/54.6
Green band 4.7/41.8 4.6/41.8 -3,6/37.1
Blue band 10.1 / 38.0 10.0/37.9 | -3.1/34.0
table 3: quality control with statistical values - transfor-
mation of two (blue, green) bands (equation 7)
(mean value / variance of differences between TC and
PTC pixel values)
class cen- | class center | classwise
ter with classwise | method
method | improvement 333
335 method 3.3.6
Red band 52/511 5.1/51.0 -3.7/44.0
Green band 4.7/41.8 4.6/41.8 -3,6 / 37.1
Blue band 10.1/38.0 10.0/37.9 |-3.1/34.0
table 4: quality control with statistical values - transfor-
mation of all (red, green,blue) bands (equation 3)
(mean value / variance of differences between TC and
PTC pixel values)
5. CONCLUDING REMARKS AND OUTLOOK
Producing pseudo-true-colour orthophotos using colour-
infrared-photographs is a possible compromise of having
excellent film material for interpretation and getting reali-
stic colour orthophotos for visualization purposes. The
presented paper showed a first approach of the determi-
nation of transformation parameters between CIR and
TC (resp. PTC) colour space.
Within the presented project the efford of producing PTC
orthophotos was very low. As the land use classification
was part of the contract with the customer, the only
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996