re
id
op BER
ul
nd
Timur Murat Celikoyan
42 43 1.1 12
3.3 A IN. 2.1
3.2 31 23 22
Figurel: Numbers of pixel neighbourhoods in the algorithm (Celikoyan and Altan, 1999)
In Fig.l. the first numbers are referring to the circle number, where the seconds to the neighbourhoods. Aiming
matching the borderline of the detail, grey value of the 1.1" neighbourhood has taken from the array, which belongs to
the image. If the difference of grey values between standing pixel and the 1.1" neighbour is in the difference border
given by the user, this pixel is matched. In the other case the same process is done with the pixel located in the 1.2™
neighbourhood. This process is used until the target pixel is found. If any targeting pixel cannot be found in a circle, the
search process goes on with the next circle. Nearby this, if a pixel is found, the next search begins with the 1" neighbour
of this circle. An important point is that, if a pixel is found with a 1". neighbour of any circle, this pixel is marked using
the 3™ neighbour of previous circle. In that way, the search process can change its rotation and every borderline from
multiple geometry can be extracted. This search process stops until any pixel could be marked using all the neighbours.
Another point is that the data size after running this algorithm is so large that it slows down the computer process speed.
In order to avoid this problem, only break points of the borderline are selected. For this purpose, gradient of this
borderline is determined in every point and the points, by which this numerical value has changed, are taken to the data.
In that way, the size of the data is minimised approx. from 1 MB to 30-40 KB. (Celikoyan and Altan, 1999)
3 EXTRACTING GEOMETRICALLY CLOSED OBJECTS
The basic process of extracting geometrically closed objects does not have a big difference from the extracting
procedure of non-closed objects. The algorithm, mentioned in Caption 2 is adapted so that it stops, when a previously
marked pixel is reached for one more time. In that way, geometrically closed objects are extracted. The handicap by this
process is to check the whole image coordinates of previously marked pixel by every search process. This difficulty
slows down the process proportionally to the running time. The big amount of pixels marked before, the slower running
speed of the algorithm. To avoid this problem partially, the object type, which will be extracted, is given by the user
before the search process. In that way, extraction process of non-closed objects does not slow down.
‘4 APPLICATIONS
As first application, damaged buildings by Dinar (1995) and Marmara (1999) Earthquakes are given. Results of these
applications can be used in Geographical Information System developed for modelling and damage analysis of
buildings. The images are taken by using Kodak DCS 200 digital camera (1524 x 1012). As it seen in Figure 2, cracks
on the damaged building have different grey values from the undamaged part. Using this difference, damages can be
extracted from digital images.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 165