mages with
Figure 12. Geocoding VNIR mosaic with marked samples.
The samples show approximately the same area on every image
from which they have been taken for monitoring radiometric
and geometric deformation during mosaicking and improvement
of geometrical relationships during geocoding. The samples I
à and III have been token into consideration because I sample was
] ; unchanged and sample III had the biggest deformation in
/ geometric and radiometric domain because that sample is the
d furthest from the first image in mosaic.
(green) and
geometrical
t has to be
garding to
ing. VNIR
id infra red
nosaics the
ing mosaics
' same area
Wea Giz Leve
ftus io
Median 57 Panik
bear CU iet
Sales HAT Count
Meter 53 Fercurtin
Pus SEE Faust GENES
Carne iew f Xaelewe !
Figure 13. Comparison between histogram of part of original
VNIR RGB image (up left — I sample, up right — III sample),
histogram of the same area, not geocoding RGB mosaic (in the
middle left — I sample, in the middle right — III sample) and
histogram of geocoding RGB mosaic (down left — I sample,
down right — III sample).
Figure 14. Comparison between histogram of part of original
TIR image (left) with histogram of geocoding TIR mosaic
(right), sample corresponds to III sample from VNIR mosaic.
SAMPLE
NUM. OF PIXELS MEAN ST.DEV. MEDIAN
VNIR
] — image 460x370 61.99 8.16 62
| — mosaic 460x370 57.52 7.16 57
1-geo mos 700x538 57.24 7.04 57
III-image 442x502 67.46 12.27 66
[[I-mosaic 398x546 64.04 12.05 63
[II-geo mos 601x695 64.07 11.47 63
TIR
III-image 128x128 193.94 22.16 196
lil-zeo mos . 512x512 192.78 22.51 195
This data show that the information on TIR image change a
little as related to original image, in spite of fact that number of
pixel have significantly increased by geocoding, and the
differences on VNIR images are more significant and their
influence can be visible on shape of histogram and MEAN and
MEDIAN values between original image and geocoding
mosaic.
S. CONCLUSION
Mosaics which were produced from digital images of
DuncanTech MS3100 and Thermovision THV 1000 are useful
information for the orientation on snapshot area and give
possibility for radiometric analysis of data on entire snapshot
area at the same time. Geometric deformation of details can be
improved by geocoding. The mean value of deviation of details
on VNIR and TIR geocoded mosaics related to DOF show that
fact, that it is possible to make joint radiometric analysis (i.e.
the classification which can be done because the pixel size is the
same on both mosaics) with biggest tolerance of results. The
deviation of VNIR geocoding mosaik regarding the model
(DOF 1:5000) are fully satisfying geometrical accuracy of DOF
1:5000 ((2) and (4)), because there aren’t any deviation larger
than 1.5 m on both axes. Standard deviations of TIR mosaic,
regarding the model (DOF 1:5000), are also within the accuracy
of DOF 1:5000 ((1) and (3)), but 4 points have y coordinate and
6 points have x coordinate deviation bigger than 1.5 m. In this
case one should take into consideration the fact that the
radiometric data VNIR of geocoded mosaic differ much from
their original copies that they have been made of, than it was the
case with TIR geocoded presentation. Base for geocoding was
DOF 1:5000 with pixel size of 0.5 m. If we had better base for
geocoding (smoller pixels of base image), geometric
transformation would give better results. For the radiometric
analysis it is necessary to find the adequate methods of
comparison.