445 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
smooth. However, it is probable the geometry of vector data 
may get worse if higher weed tolerance is given (Eker, 2006). 
3. TEST STUDIES 
Test studies were carried out with the program developed in 
two groups in order to determine the usefulness and 
effectiveness of this new approach exposed. 
In the first group test study a 1:16000 scale black/white digital 
aerial photograph was used. Thus it was investigated whether 
the software would be used or not to extract some features in 
large scaled mapping. 
In the second group study Another application with 1:5000 
scaled two ortho images which have 0.50m resolution. These 
ortho images are generated from 1:16000 scaled color aerial 
photographs. In this test area, an accuracy test is also carried 
out to find the accuracy of the developed method. 
3.1 Test 1 
In the first test work that a 1:16000 scaled white/black aerial 
photo which was scanned at 20 microns was used. This image 
includes a village center. Roads were selected as the features 
that were extracted semi-automatically from the image. It is 
shown in the Figure 1 that vector data which were obtained 
from the result of feature extraction studies. Roads were 
vectorized from their edge lines. The roads’ surfaces are not 
homogeneous. Because of this in this image the extraction of 
the roads were a hard work. In this image, if we selected a big 
tolerance value the extraction extended out of the roads. On the 
other hand if we selected small tolerance values we saw that we 
had to make small propogations and select a lot of starting 
points. This increased the time of the extraction. 
(a) • (b) 
(c) 
Figure 1. (a) The original image (b) Mask image of the 
extracted roads (c) Extracted vectors. 
3.2 Test 2 
It is thought that this software can be used in producing maps 
and in collecting the vector data for GIS. Another application 
with 1:5000 scaled two ortho images which had 0.5m 
resolution of Ayazaga Campus of istanbul Technical 
University. These ortho images were generated from 1:16000 
scaled color aerial photographs. In this test area, an accuracy 
test was also carried out to find out the accuracy of the 
developed method. In this accuracy test, vector data of the roads 
and buildings were collected semi-automatically with the 
developed sofware (Figure 2) and also manually with an 
experienced operator (Figure 3). The data collected by the 
operator were assumed as the correct ones and they were 
compared with the others collected by the software. The 
accuracy test was carried out in two groups. In the first group, 
roads were used and on 422 road check points were measured, 
In the second group, buildings were used and 281 check points 
were measured. The results are shown in Table 1. 
é **» !<-*» V, *«. 
Figure 2. Semi-automatically extracted features. 
Figure 3. Manually extracted features by the operator.