A STUDY ON THE REPRESENTATION AND 
INTERPRETATION OF REMOTELY SENSED IMAGERY 
Li Deren 
Guan Zequn 
Department of Photogrammetry &. Remote Sensing 
Wuhan Technical University of Surveying and Mapping 
Wuhan 
ABSTRACT 
For past many years,many reseatchers have devoted 
effort the of 
representation. However, no recent study has been 
considerable to problem image 
undertaken for an interpretation of remotely sensed 
image based on the combination of multiform 
representations under conditions where a computer 
system can interpret an image by incorporating multi 
— information overlap analysis in the traditional 
representation of image understanding. In this paper, 
we describe an multilevel representation tree composed 
of synthetic land unit map ,land unit relatability list, 
complexes labeling, correlation list and region relation 
graphs. This representation tree allows one to interpret 
image in terms of remotely sensed image and non — 
that may provide 
remotely sensed digital maps 
geometric, topological, overlapping and attributive 
information in a form suitable for image 
interpretation. We present an interpretation for 
remotely sensed image by using multilevel 
representation tree and give also a number of 
experimental results. 
KEY WORDS; 
Multilevel 
Representation, Interpretation, 
representation tree, Synthetic land unit 
map , Complexes labeling , Correlation list. 
1. INTRODUCTION 
The problem addressed in this paper is that of how 
effectively to interpret an object in an image based on 
the combining of remotely sensed images and non — 
remotely sensed digital maps that are in multiform 
878 
China 
representations. 
For many years, much of the research on image 
recognition has involved various representatations 
such as boundary representations, constructive solid 
geometry representations, sweep representations , and 
etc... Most of 
methods are feasible for acquiring the geometric 
decomposition representations these 
information from the object image (Andrew, 1989). 
We submit that the above — mentioned representations 
provided for image recognition all have two grave 
defects. The first defect is that some information ,such 
as the between and their 
relationships objects 
environmental factors, which is very usuful for 
remotely sensed image interpretation hasn? t been 
reflected. The second defect is that it is impossible to 
obtain a suitable correspondence between different 
data that may be remotely sensed images or digital 
maps. This is because the representations are aimed at 
representing regular geometric form. 
It has been shown that geographic information is 
important to remotely sensed image interpretation, 
because the data Provide information on the spatial 
distribution of important factors such as rivers, roads, 
towns etc.. To use the geographic information and 
overcome the shortcomings as stated above, in this 
paper we describe a multilevel representation tree that 
have two principal branches. One of them is 
composed of various digital maps, synthetic land unit 
list etc. , which 
map and land unit relatability 
originate from multi information overlap 
analysis. The other is associated with complexes