SEMI-AUTOMATIC EXTRACTION OF FEATURES
FROM DIGITAL IMAGERY
0. Eker a *, D. Z. Seker b
d General Command of Mapping, Photogrammetry Department, 06100 Dikimevi Ankara, Turkey - oeker@hgk.mil.tr
b ITU, Civil Engineering Faculty, 34469 Maslak Istanbul, Turkey - seker@itu.edu.tr
KEY WORDS: Image Segmentation, Semi-Automatic, Feature Extraction, Fast-Marching, Level-Set
ABSTRACT:
Aerial photographs have been evaluated manually by the operators for a long time for the extraction of the vector data. The
development of computer technology and digital image processing technologies provide to perform these extraction processes
automatically or semi-automatically. Image segmentation can be used for automatic and semi-automatic feature extraction and
classification of images. In recent years, image segmentation and the front propagation of the segments have been carried out
successfully by the Level Set and Fast Marching methods. In this study, a semi-automatic line extraction method, based on the
segmentation of the images using color-differences of the pixels and the propagation of the fronts by the Level Set algorithms, is
developed. An object oriented application software is also developed to test the capabilities of the developed method. Some semi
automatic feature extraction applications are made by the help of the developed software using a 1:16000 scale black/white aerial
photograph for determining the capabilities of this method. Another application with 1:5000 scaled two ortho images which have
0.50m resolution. These ortho images are generated from 1:16000 scale color aerial photographs. In this test area, an accuracy test is
also carried out to find the accuracy of the developed method. The accuracy test is carried out in two groups. In the first group, on
422 road check points, measurements are made and the root square mean is found as ±0.66m. In the second group, buildings are used
and 281 check points are measured and the root mean square of this group is found as ±0.46m.
1. INTRODUCTION
Aerial photographs have been evaluated manually by the
operators for a long time for the extraction of the vector data.
Computer technology and digital image processing technologies
have been developed and this development provides to perform
these extraction processes automatically or semi-automatically.
The aim of making the processes automatic is to increase the
speed of collecting the data and to reduce the cost. Automatic
feature extraction studies are firstly motivated to carry out the
extraction of roads from digital images because roads have
characteristic attributes like width, surface type and geometrical
shape which can be modelled more easy than the others.
These studies are showed that the resolution of the images has a
very important role in the automatic and semi-automatic
extraction of the roads. The developed methods can be
classified in three groups: road extraction in low resolution,
road extraction in high resolution and multi-resolution road
extraction. Most known methods are based on the road tracing
and the snakes algorithms.
Another method of automatic and semi-automatic feature
extraction and classification of images is the image
segmentation. Image segmentation is commonly used the
interpretation of the medical imagery like mr-scans
(Adalsteinsson and Sethian, 1995).
In recent years, image segmentation and the font propogation of
the segments have been carried out successfully by the Level
Set and Fast Marching methods (Malladi, 1994).
The goal of this study is to develop a semi-automatic extraction
method of the line and area features like roads, rivers and lakes
from digital aerial photographs and updating of these features in
Geographic Information Systems (GIS) and also to develop an
application software to test the capabilities of this method.
According to this goal, a semi-automatic line extraction method,
based on the segmentation of the images using color-differences
of the pixels and the propogation of fronts by the Level Set
algorithms, is developed. An object oriented application
software is also developed using Borland C++ and Visual C++.
2. SEMI - AUTOMATIC FEATURE EXTRACTION
APPROACH AND DEVELOPED SOFTWARE
The approach that has been developed for semi-automatic
feature extraction has been based on level set and image
segmentation algorithms. According to this method, three
problems had to be solved. The first one was how the algorithm
would be started. This problem was solved by marking any
point (pixel) on the feature wanted to be digitized by the
operator. Thus, level set algorithm would start to work from the
point that operator chooses. This solution brought out the
approach to be semi-automatic.
Second problem was which criteria would be taken in whether
the digitizing feature process would progress or not. This
problem was solved by making use of color values of each pixel.
The color value of marked point or determined neighborhood
level, color values of neighboring pixels and computed color
value by getting average, is compared with color value of
neighboring pixels and if the color difference is within the
Corresponding author.