168 
A HIERARCHICAL TERRAIN INTERPRETATION SYSTEM 
USING ’PIXEL SWAPPING’ METHOD 
Joji Iisaka 
Canada Centre for Remote Sensing 
1547 Merivale Rd. Nepean, Ontario, Canada 
Wendy Russell 
Intera Technologies Ltd. 
Ottawa, Ontario, Canada 
ABSTRACT 
Digital image processing is very popular for extracting terrain information from remotely sensed data. At 
present, most image analysis methods are still limited to the processing of images on a pixel-by-pixel basis while 
the human interpretation of images is mostly based on structural spatial information with a high level of context 
and semantics. This paper briefly describes the concept of the microcomputer-based terrain understanding 
system which employs hierarchical image interpretation steps using the ’pixel swapping’ method. 
The ’pixel swapping’ method provides a unified image analysis method for spectral and spatial features in 
images. Image interpretation steps are divided hierarchically into a bottom-up and top-down approach. Each 
involves a few steps. With the bottom-up approach, an image is analyzed first by traditional low-level 
processing methods using numerical pixel values. The results of the numeric interpretation are interpreted in 
a more symbolic way by applying the ’pixel swapping’ method of spatial interpretation. The symbolized images 
are interpreted based on the type of the attributes, constraints of the attribute values, and the spatial relations 
between various objects in images. 
On the other hand, the targeted object in an image is divided into its constituent components and their 
attributes under the top-down approach. The system gives the control to the lower levels to evaluate required 
attribute values, if they have not been evaluated before, and compares the estimated attribute values with 
existing knowledge of the attributes in a data base. 
The system employs an LSI-based image processor (IP9506) to accelerate low-level image processing and to 
facilitate the implementation of pixel swapping.