GENERALIZATION OF IMAGE DATA TO GIS POLYGONS
FOR CHANGE DETECTION AND DATA BASE REVISION
Katarina Johnsson, PhD
Dept. of Geodesy and Photogrammetry
Royal Institute of Technology, KTH
S - 100 44 Stockholm
Sweden
Commission IV, Working Group 3
KEY WORDS: map revision, forest maps,.environmental monitoring, remote sensing
ABSTRACT: Semi-automatic revision of geographic data bases from satellite imagery requires methods to perform
change detection between map and image data. Most geographic databases are stored in vector-based geographic
information systems, while analysis of satellite data requires raster GIS. It is argued that image data must be
generalized into one or several polygon attributes to enable change detection between map and image data, and that
change detection should take place in an attribute data base environment, rather than in a raster image environment. A
framework for image generalization functions is presented, based on descriptive statistics. Statistical measures to
describe central tendency, dispersion and spectral heterogeneity are discussed for quantitative and qualitative image
data respectively. A case study is presented, which describes how forest change detection can be performed by
regression between polygon attributes derived from map and image data of different dates. High detection accuracy is
achieved for logged forest stands, confusion with objects of similar spectral characteristics is avoided, and the
influence from natural spectral variation is minimized.
for soil and forest maps. In these cases, the revision
process has two distinct steps: 1) detection of changes
1. INTRODUCTION with respect to database attribute(s), and 2) adjustment
of geometric properties, e.g. by drawing new boundaries
There is a need for efficient methods to revise digital and deleting old. This paper discusses the first step.
geographic data bases. This paper is concerned with
revision of small-scale geographic data bases Map revision from satellite data requires that change
(>1:20,000) for natural resources management, and detection takes place from map to image data. This can
specifically with revision of map polygon features. be done manually, by on-screen image interpretation and
Satellite imagery provides an adequate data source for map overlay. However, operational revision of large
revision for geographic databases at this scale. databases require the revision process to be, at least,
semi-automated. Several issues need to be addressed
A geographic data base will normally contain one or before semi-automatic database revision becomes
several layers of digitized maps. Most geographic data reality, for example differences in data formats and in
bases have been created and are being maintained in levels of abstraction between database and image data.
vector-based Geographic Information Systems (GIS),
where map features are represented either as points,
lines or polygons. Polygon features are especially 2. CHANGE DETECTION FROM MAP TO IMAGE
common in geographic databases for natural resources
management, for example forest stands, animal habitats, Since integrated vector/raster analysis is currently not
soil compartments, geomorphological terrain possible, one of the data sources must be processed to
components, and watersheds. fit the other, before map-to-image change detection can
take place. This is not only a question of format
Satellite imagery is stored and processed in raster based conversion, but also a question of data abstraction.
GIS or image analysis systems (IAS). Traditionally, Maps are highly abstracted representations of the real
change detection has been focussed on pixelwise world, while image data is a lower form of information
analysis between images of different dates (Singh, (Ehlers et al., 1991; Goodchild, 1989).
1989).
There are several reasons for why image data should be
Revision of geographical databases can take place generalized to map polygons, rather than converting the
either with respect to map attributes or with respect to vector map to raster data and performing pixelwise
geometric properties. For natural resources databases, change detection (the traditional image analysis
geometry and attributes are often interconnected, since approach):
the boundaries of the map polygons are drawn based
their attributes (Veregin, 1989). This is for example true ° The image data becomes immediately comparable
to other polygon attributes in the attribute
Current affiliation: T-Kartor Sweden AB, Box 5097, database.
S-291 05 Kristianstad, Sweden
email: katarina.johnsson @t-kartor.se
384
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
Thes
the ir
relev
level
imag
attrib
(vec
previ
from
et al.
3.
As €
gene
in ra
area
(199
Imag
regio
pixel:
to re
Conc
imag
base
quar
meas
has |
pixel
Imag
class
3.1
func
Gene
inclu: