The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
446 
Mean 
RMS 
Min.-Max. Errors 
Roads 
0.93 m. 
±0.66 m. 
0.10 m. - 4.18 m. 
Buildings 
0.71 m. 
±0.46 m. 
0.03 m. - 2.29m. 
Table 1. Results of the accuracy test. 
4 CONCLUSIONS 
As the results of the applications and tests, it can be said that 
the accuracy of this developed method is ±1 pixel size of the 
used imagery. It can be used correctly for producing maps and 
collecting vector data for GIS. Especially for lakes, rivers and 
buildings can be collected very efficiently. Different 
classifications and segmentations, which an operator’s can not 
see, can be made also with the adjusting of the tolerance value. 
Roads which have good quality can be vectorized from their 
center lines and/or boundaries according to the scale of the 
image used. 
Some weak sides of this developed method and software are 
also found out. These are: 
• Especially on big scale aerial photographs, the obstacles on 
the features, as trees, cars and shadows, effects the extraction of 
the features negatively. Effects of this factor are reduced 
whether the scale of the image gets smaller. 
• If the tolerance value is not be adjusted to the correct values, 
wrong features can be extracted. 
• When a big size image is used, the software gives back 
some errors because the size of the arrays are directly 
proportional to the number of the pixels. • 
• The quality, contrast and noise of the image effect the 
feature extraction process. 
• The surface attributes of the features also effect the success 
degree of the feature extraction. 
If the noise and the contrast of the images are eliminated by the 
image process algorithms like edge detection algorithms and 
filters as anisothropic diffusion and the blanks that are 
generated by the obstacles on the feature can be enterpolated by 
the different kinds of entepolation methods, more good results 
can be achieved by the developed method and the software. 
Also, for the image segmentation different types of 
segmentation like snakes, instead of color difference and for big 
size images pyramid levels can be used to increase the success 
degree of this method. 
REFERENCES 
Adalsteinsson, D., Sethian, J.A., 1995. Fast Level Set Method 
for Propagating Interfaces, Journal of Comp.Phys., 118, pp 269- 
277. 
Eker, O., 2006. Semi-Auotmatic Extraction of Linear Feature 
From Aerial Photograps, Ph.D. Thesis, ITU, istanbul. 
ESRI, 1997. ARC/INFO User’s Guide Cell-Based Modelling 
With GRID, Redlands, USA. 
Malladi, R., Sethian, J.A. ve Vemuri, B.C., 1994. Evolutionary 
Fronts for Topology-Independent Shape Modeling and 
Recovery, Proceedings of Third European Conference on 
Computer Vision, Stockholm, Sweden, Lecture Notes in 
Computer Science, 800, pp 3-13. 
Sethian, J.A., 1998. Fast Marching Methods and Level Set 
Methods for Propogating Interfaces, von Karman Institute 
Lecture Series, Computational Fluid Mechanics, Belgium. 
URL ( Raster to vector conversion open code), 
http://www.xmailserver.org/davide.html, 28 February 2006.