Istanbul 2004 
  
SEMI AUTOMATIC DIGITIZING OF CONTOURS FROM 1:25000 SCALED MAPS 
C. Helvaci* *, B. Bayram * 
“ YTU, Civil Engineering Faculty, 34349 Yıldız, Besiktas Istanbul, Turkey — (chelvaci, bayram)@yildiz.edu.tr 
Commission III, WG 11/4 
KEY WORDS: Digitization, Extraction, Rectification, Algorithms, Raster, Vector 
ABSTRACT 
It is an expensive and time consuming task to digitize 1:25000 scaled maps which are made in the past with analogue methods. The 
development in digital image processing depending on the development of hardware technology made this process less expensive 
and faster. This study is based on this reality. In this thesis a new algorithm which aims to decrease human work time by making the 
work carried out to digitize 1:25000 scaled maps easily and less time consuming was developed. The first step of the process is 
interactively rectification of the map. The grid network is used for rectification. Scanned maps at the scale of 1:25000 and the 
training area on this map which is given by the operator that is showing the elevation contour pixels are used as input data. The other 
elevation contour pixels are determined automatically. Elevation contours in vector format are created from the elevation contours in 
raster format. Produced vectors may contain errors depended on the topographic condition of the map area. Elevation contours with 
XY coordinates are generated after the manual editing. The Z values of these contours are given by the operator before the end of 
the process. Finally, produced data is saved in ASCII format which is supported by common CAD, GIS and engineering software. 
The output file contains X, Y, Z values delimited with comma (“,”). 
1. INTRODUCTION 
Automatic information extraction from map images is not a 
trivial task since objects in the charts must be interpreted from 
their spatial characteristics such as shape, scale and contours 
(Esposito F., et al.1998). Vectorization, also known as raster to 
vector conversion, is a process that finds the vectors — straight 
line segments — from the raster images (Wenyin, L. Et 
al.,1999). The contour-pixels detection consists of edge 
detection, thinning, and linking operations. These are window 
operations, in which the output at a pixel is based on the value 
of the input pixel and the value of its neighboring pixels (Chung 
Y, et al., 1995). 
In Turkey many of the 1: 25 000 scaled maps are analogue. 
Geodetic and planning applications need these maps in digital 
format. Conventional digitizing process of 1: 25 000 scaled 
maps according to the topography of the area takes a very long 
time; sometimes a month. When this process is considered for 
each of the governmental and public institutions, the abundance 
of cost and time spent can be imagined. On the other hand, 
accelerated production of digital terrain models gives a 
possibility to design the projects in minimal time and cost. 
Algorithms like these, will avoid the errors that occur because 
of the human factor. 
The aim of this study is to minimize the digitizing time of 
1: 25 000 scaled maps. From the results of the study, it can be 
declared that this process reduced the digitizing time to a period 
between 4 hours and a week. Naturally, this time depends on 
the topological structure of the area and deformation of the 
map. 
This software can be used with 200 dpi resolution scanned 
maps. 
  
2. THE SEMI-AUTOMATIC DIGITIZING ALGORITHM 
OF CONTOURS 
Semi - automatically recognition of the contours in 1:25000 
scaled scanned maps is made by the process below: 
2.1 Rectification 
2-D Affine transformation method is used to rectify the 
1: 25 000 scaled map. 
2.2. Collecting the Training Samples 
The method works with 24 bit true color image. To collect 
training sample contours, the points below should be checked: 
1. The contours that have the same color but different 
gray values should be grouped. 
2. The training samples should be in the middle axis of 
the contours. 
3. The contours which have the same color but different 
brightness should be grouped. 
To have accurate measurement results, the collected samples 
have to be distributed normally. There is no need to make any 
statistical adaptation tests to control the distribution of pixels 
selected by the user. Instead, the algorithm in section 2.3.1. is 
able to demonstrate the contours in whole image depending on 
the training measurement of users. With this method, the most 
accurate group is chosen interactively. 
2.3. The Automatic Recognition of the Contours 
The linear approximation consist of contour tracing and 
approximation operations (Chung Y, et al., 1995) . Algorithm of 
deformable model method needs premise measurements for 
* Corresponding author. This is useful to know for communication with the appropriate person in cases with more than one author.