POLYGON BASED ANALYSIS OF REMOTELY SENSED IMAGES IN AN INTEGRATED GEOGRAPHIC 
INFORMATION SYSTEM 
Eugene Derényi, Hon. Research Professor 
Mustafa Türker, Research Associate 
Department of Geodesy and Geomatics Engineering 
University of New Brunswick 
Fredericton, N.B. Canada E3B 5A3 
Commission IV, Working Group 6 
KEY WORDS: Change Detection, Monitoring, Updating, Database, Knowledge Base, Image Classification, Image 
Statistics, Raster Vector Integration. 
ABSTRACT: 
Two schemes were developed and successfully implemented for monitoring changes and prevailing land cover conditions, 
within geographic boundaries stored in a GIS, by digital image analysis. In the first scheme, a standard pixel-by-pixel image 
classification is performed, one theme at a time, covering the entire area of interest. A polygon-by-polygon assessment of 
the results follows to separate those polygons which deviate from the apriori expectations by more than a preset threshold 
level. These polygons are then subjected to further examination. In the second scheme, image classification is bypassed 
entirely and only image statistics are generated polygon-by-polygon. A comparison of these statistics with those 
representing normal, expected conditions identify those polygons where significant changes occurred. Both schemes are 
  
demonstrated by an example. 
The above approach to monitoring changes overcomes the problems of multi-classification and misclassification due to 
overlaps of spectral signatures, which often plagues traditional image classification. In addition, it provides for an 
immediate update of the data base in the GIS. 
1. INTRODUCTION 
The Earth is a planet with finite resources. As the 
population continues to grow, a wise and prudent 
management of these resources is becoming increasingly 
important. Such management is best achieved if accurate 
and up-to-date inventories are at the disposal of decision 
makers. More and more of the information needed for this 
purpose is generated by remote sensing. An image is worth 
a thousand words. 
Remote sensing is, however, not an end in itself. 
Classified images displayed on a monitor or as multi-colour 
plots are end product of limited value. The realistic 
approach is to store the information extracted from images 
in a geographic information system (GIS) where it can be 
merged with other information pertinent to resource 
management. This is, of course, becoming more and more 
the norm. The closer GIS and remote sensing become 
integrated the greater the benefits derived by resource 
managers become. 
The flow of information in the opposite direction is 
equally important and beneficial. GISs now hold vast 
volume of well-organized information which could serve as a 
knowledge base to facilitate the analysis of new remotely 
sensed data. This advantage has, of course, been recognized 
for some time and topographic, geophysical, geological, 
soil, etc. data are often utilized as additional raster layers in 
image classification. Administrative, land use, etc. 
boundaries are also being used for segmentation and 
stratification of images. In addition to the geographic. 
features and topological information, GISs also contain 
attribute data which are stored in a relational data base 
management system (DBMS) interfaced with the GIS, and 
can be queried. Such descriptive information about features 
is, however, rarely utilized in image analysis since their 
alpha-numeric storage format is not compatible with the 
raster organization of digital images, and most digital image 
analysis systems are not interfaced with relational database 
management systems. Therefore, a universal GIS (UGIS) is 
needed where both vector based geographic data and raster 
based continuous tone image data can be processed, analyzed 
and manipulated and which is interfaced with a DBMS. The 
Computer Aided Resource Information System (CARIS) used 
in this project, has most features of such a facility [Derenyi 
and Pollck, 1990; Derenyi, 1991]. In a UGIS the per pixel 
image classification can be replaced by polygon specific 
image analysis, and apriori knowledge, stored in the DBMS, 
can be incorporated in the decision making. The results can 
directly be stored in the database. 
2. THE CONCEPT 
The polygon specific image analysis scheme presented 
here has been developed for detecting and monitoring 
changes within existing geographic boundaries, which are 
stored as polygons with associated attributes in a GIS. Final 
class assignment is made polygon-by-polygon, based on 
information extracted from the raster data layers and on the 
polygon attributes stored in the DBMS. The raster layers 
may include image and non-image data. 
Two methods have been employed for analyzing the 
image data, and to arrive at the final class assignment of 
individual polygons: 
1. modified per-pixel image classification, and 
2. decision by polygon specific statistics generation. 
In the first method, supervised classification is performed 
covering the entire project area. This classification is done 
one theme at a time, and each class is stored as a separate 
layer. Training samples are selected with care as usual, but 
multiclassification of pixels caused by spectral overlaps i$ 
of no major concern since the output of each class is clearly 
212 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996 
ser 
tha 
an 
po! 
on 
cla 
cla 
for 
wi 
the 
lab 
dec 
po! 
ex 
avé 
pei 
as 
rer 
ex| 
on 
dis 
pol 
nol 
fur 
iss, 
pix 
sta 
thc 
the 
Cal 
Spe