FEATURE EXTRACTION COMPARISON OF IMAGE ANALYSIS SYSTEMS
AND GEOGRAPHIC INFORMATION SYSTEMS
J Gairns, Intera Information Technologies, Canada
T Taylor, DIPIX Technologies Incorporated, Canada
ABSTRACT
Today, user demands and improvements in Information Systems are bringing together data formats that were not previously
physically linked. With the advent of Geographic Information Systems (GIS), it was realized that data from a wide variety of
sources could be used in a complimentary fashion. One example is remotely sensed imagery in a GIS environment. A GIS
coverage has a unique feature that remotely sensed imagery does not. A database. A database is an integral component
of a GIS, but it requires extensive management in order that useful information may be stored and manipulated. Remotely
sensed data offers an elegant supplement to such a database, in the form of extensive spatial information. Information from
a raster image can be extracted automatically, with or without operator supervision. By using an image as a data source and
the GIS vectors as boundaries to delimit that data, the two environments can offer more functionality than either alone.
BERT BESSER: SSR EINE DR
There are concerns regarding the marriage of these two data sources. If the GIS data, once brought together with the raster
data is not properly registered, problems can arise. For example, data that is not adequately geo-referenced is hardly useful
at all, and results derived from such can easily be highly deceiving.
The raster and vector environments can offer a great deal of information exchange to each other. In fact, the existence of
either data type enhances the information content of the other. Given this perfect marriage, one must consider the
consequences of bringing one type of data into the realm of the other. What happens to raster data when it becomes
vectorized and conversely, what happens to rasterized vector data. Is there perfect registration, or is the registration of these
complimentary data not as straightforward ? An ARC/INFO vector polygon coverage of water boundaries was integrated with
A an ARIES format NOAA AVHRR Local Area Coverage (LAC) image of the corresponding geographic area. The GIS polygons
ial were rasterized and visually compared to the existing NOAA sub-scene. A Feature Extraction technique was performed on
on the NOAA sub-scene and the resulting vectors were compared against the original GIS coverage using a simple visual
re, comparison.
in
m- Keywords: Feature Extraction, GIS/IAS Integration, Accuracy
of
st,
Ip- 1.0 INTRODUCTION
rd, Most existing research into comparative or relative feature
Traditional Image Analysis Systems (IAS) offer an ideal compliment extraction deals specifically with the use of some sort of interactive
37. to GIS data extraction, manipulation and archiving functionality. component to the procedure (Schowengerdt and Pries (1988),
m- The extraction of image statistics using a GIS overlay is an obvious Zelek (1990), O'Brien (1991). These are perfectly valid
3 - benefit. For example, an operator can automatically select training approaches, but a user may not always have such intimate
areas by using the functionality of the GIS and querying the pixels knowledge of a study site which reduces the potential efficiency of
ng that fall within a polygon. This information could easily be stored the extraction technique. What does one do in this case ? The
10- in a database, and subsequently manipulated as a database answer points to an automatic approach. Work in this area is still
: attribute. very much in the research phase, although it is approaching an
et- operational level.
pp. Given that spatial data has an extremely high information content
for a relatively low cost, it is desirable to integrate spatial data with À particularly important concept in the field of feature extraction is
tás a topological database, such as is inherent in a GIS. Spatial data how an algorithm actually recognizes an edge or boundary. The
le offers vast quantities of information, but one must consider what procedure for locating linear features is very similar to that of
happens when spatial data is brought together with other data locating spatial features. Both have edges, which can be thought
for types. The purpose of this paper is to explore the implications of of as a "contrast amongst distinct features in the image” (Zelek,
merging traditionally detached sources of information via 1990). Characteristics of an edge such as pattern, size, shape and
10- automated, or semi-automated procedures, in particular, feature colour are important elements in the recognition of the contrast that
Dp. extraction. delineates the edge of a particular feature. It is difficult to quantify
these characteristics, however a qualitative approach can prove
Ty. There is a desirable effect when data from a GIS is merged with useful as a tool for comparison in this case.
remotely sensed imagery. This serendipitous effect is information
synergism. Information synergism is the overall increase in
information content of a system, exceeding that of the individual
data components. By modelling the various data types in a single
environment, information that was not previously obvious becomes
evident.
365
2.0 DATA DESCRIPTION
Practically speaking, most remotely sensed imagery could be used
in a study such as this, however NOAA AVHRR was chosen for the
task. The choice to use this data was based on the availability of
