ing
cts.
nge
ion
295
of
ical
Alan Forghani
SEMI-AUTOMATIC DETECTION AND ENHANCEMENT OF LINEAR FEATURES TO UPDATE
GIS FILES
Alan FORGHANI
School of Geoinformatics, Planning & Building
University of South Australia, City East Campus
PO Box 2471
Adelaide, SA 5001
Australia
Ph: +61-8-83021871
Fax: +61-9-83022252
E-mail: alan forghani 9 hotmail.com
Working Group IV/2
KEYWORDS: Edge Detection, Thresholding, Mathematical Morphology, Semi-Automatic Linear Feature
Detection GIS, Mapping, Aerial Photography.
ABSTRACT
This paper describes a program developed to allow semi-automatic detection and enhancement of linear features in
aerial photography. The program employs three different edge detectors namely Sobel, Deriche and Canny as well as
morphological operations. The effectiveness of each of these operators and mathematical morphology is compared.
Classification accuracy evaluation demonstrated that the Canny edge filter gave best results among these algorithms.
The program is able to extract roads, field boundaries, and rivers to update GIS databases.
1 INTRODUCTION
Mapping of road networks is a topic of interest to both the Geographic Information Systems (GIS) and remote sensing
(RS) communities. The problem of keeping road network files up to date is most acute in the urban fringe of major
urban areas where development processes are most concentrated.
A program, called the Interactive Linear Feature Detection Program (ILFDP), was developed for semi-automatic linear
feature detection using different edge detectors and morphological operations (Forghani, 1997a). Three different types
of spatial filters were employed inthe program: a) noise removal filters b) edge detectors (Sobel, Canny and Deriche)
and thresholding, and c) mathematical morphologic transformations.
Aerial images were processed using ILFDP implemented in MATLAB. ILFDP is able to extract edges of roads, field
boundaries, and rivers. The extracted edges can be used to update GIS databases.
The purpose of the program is to bring together a set of routines into an easy interactive 'suite' of programs to enable the
user to optimise edge detection. Different edge detection filters and thresholding values can be compared in order to
find out which filter and process is most effective and efficient for particular imagery.
2 BACKGROUND
Image processing techniques include certain types of image segmentation such as edge enhancement, edge detection
and mathematical morphology.
21 Edge Detection
Edge detection refers to the identification in an image of edges such as object boundaries, or abrupt changes in surface
orientation and material characteristics (Van Der Hejin, 1995). Trade-offs between edge detectability, noise sensitivity,
and computational efficiency are often involved in selecting a suitable edge detector for a given application. Edge
detectors generally suffer from weaknesses not only in sensitivity to noise, but also in poor performance near corners of
structures (McKeown and Zlotnick, 1990). A successful edge detector for image segmentation depends upon a number
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 289