cM 
isers, 
Lon 
in more 
jS. The 
ountry 
3,- about 
7 MKF-6 
bands 
to and 
space 
> user 
imeras 
ight 
20 cn, 
> pho- 
iluation. 
.n model 
.On pro- 
resolu- 
' even 
ling to 
Tol- 
luick 
ent 
trans- 
ster 
lotogra- 
volume 
11ly 
rmation 
editi- 
orma- 
16 
cial 
'S spe- 
mathe- 
rtogra- 
hem 
logue 
llows 
f an 
tape 
nit, 
e com- 
appli- 
ication 
1 
erty of 
same 
P» 
  
building. It will be used as host computer for future extensions 
related to interactive digital image analysis terminals. 
Obviously, our present image processing capabilities are not suf- 
ficient without any special devices like high resolution raster : 
scan display with feedback loop processor, a larger capacity direct 
access storage media and interactive tools like data tablet, 
trackball. So, a new, image computer and display terminal will be 
added to the present PDP 11/40 configuration, which will enable 
real time interactive digital image analysis with better cost/per- 
formance ratio. : = 
As far as the analogue instrumentation is concerned a large scale 
of stereophotogrammetric devices /Stereometrograph, Stecometer, 
Topocart Orthophot etc/ and other units.specially applicable for 
remote sensing tasks like large format optical rectifier, point 
marker, interpetoscope, Stereo Facet Plotter etc are at our dis- 
posal. For a better and faster service to our customers, the RSD 
will recently install a b/w and colour photographic laboratory. 
The completion is scheduled for the next year. 
Let's speak about the software package developed by our Digital 
Image Processing Lab. Emphasized first of all for small-scale the- 
matic mapping using Landsat imageries, the following routines are 
already in operational use: 
Image processing utilities:  I/O routines, conversions, file main- 
tenance, reference /incl.training/ data handling.. 
Image preprocessing routines 
- pemoving the stripping which results from different responses 
of Landsat sensors. /Fig.3./ The applied statistical correction 
models were first introduced by the japanese Earth Observation 
Center and by the M.I.T. Artifical Intelligence Lab, 
- various arithmetic relations between two registered images 
- 3x3 moving window filters for smoothing and edge detecting /Fig. 
37, 
- special filters /pseudoplastic representation/, 
- contrast streching /as used by the Technicolor Graphic Service 
at the EROS Center 
- windowing /selecting/ training areas, subimages /presently in 
quadrangle form only/, 
- calculating statistics for training areas, displaying one and 
two-dimensional histograms and scatterograms, 
- Karhunen-Loewe principal component transformation of an image, 
- simple correction for geometrical incompatibility of Landsat/ 
Colormation pixel sizes, 
- geometric correction and registration. The routine registers 
one scene to another or to the National Unitary Projection grid 
by means of ground control points which are easily identificable 
on the images. The transformation procedure based on the least 
squares method. In another mode the program waits for the 
precalculated transformation parameters. In both cases rotation, 
shifting, flexible scale selection is supported by the software. 
At the resampling the nearest neighbour and cubic convolution 
methods are used. Q 
Classification and related routines: 
- unsupervised classification /clustering/. The algorithm delinea- 
tes automatically classes by looking for clusters in the N-di- 
mensional feature space. The widely used ISODATA method is imp- 
lemented. i 
99 
Fr ET 4 à 
" ae i E SEER ON, DEN CASA me aa P A on