ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001 
178 
A STUDY ON THE EXTRACTION OF DEM FROM SINGLE SAR IMAGE 
Mingsheng LIAO 1 , Jie YANG 2 , Hui LIN 1 
1 Joint Lab. for Geoinformation Science, Chinese University of Hong Kong 
Shatin, N. T., Hong Kong Email: msliao@cuhk.edu.hk 
2 National Lab. for Information Eng. in Surveying, Mapping and Remote Sensing, Wuhan University 
Luoyu Road 129, Wuhan, 430072, China 
KEY WORDS 
SAR image, radarclinometry, digital elevation model, shape-from-shading 
ABSTRACT 
Topographic information is important in many geographic applications. Synthetic Aperture Radar observes the earth with side-looking 
imaging mode. Its image is very sensitive to the terrain shape. The tittle undulation of the terrain may induce the change of the image 
gray distribution and/or the texture characteristics. In this paper, the radarclinometry for extracting the earth elevation from only single 
SAR image is discussed, which is based on the shape-from-shading principle developed in computer vision. The post-processing of the 
elevation data is further investigated. The promising experiment result is presented. 
1. Introduction 
Topographic information is important in many geographic 
applications. Synthetic Aperture Radar (SAR) is a kind of 
imaging sensor capable of producing high-resolution images in 
all weather and all-day conditions. But it usually needs image 
pair or more to generate DEM from SAR imagery, such as 
interferometric SAR (InSAR) and radargrammetry. 
InSAR is based on the analysis of the phase difference 
between the backscatter signals received by two antennas. If the 
conditions in spatial and temporal baseline are satisfied, INSAR 
may allow production of high resolution digital elevation model 
(DEM) and detection of small earth motion. It is a highlight 
approach for its potential in accuracy and efficiency. But it is 
strict for the parameters such as the look angle, the baseline. 
The procedures for data processing are still complex. For 
example, it is very difficult to acquire the high coherence image 
pair in the area densely covered with vegetation. 
Another alternative one is called radargrammetry, a conventional 
stereoscopic method, with which the elevation is derived through 
parallax measurement from two overlapping radar images. With 
the launch of Canada's RADARSAT and ESA’s ENVISAT/ ASAR 
the generation of DEM from radargrammetry has once more 
become a remarkable R&D topic. But it still stays problems 
concerning the auto-matching of SAR images which result is 
strongly disturbed by the speckle noise. 
In this paper, the radarclinometry for extracting the earth 
elevation from only single SAR image is investigated, which is 
based on the shape-from-shading principle developed in 
computer vision. 
It is well-known that SAR scans the earth with the 
side-looking imaging mode and its image is very sensitive to the 
terrain shape. The tittle undulation of the terrain may induce the 
change of the image gray distribution and/or the texture 
characteristics. Radarclinometry entirely considers radiometric 
information of single SAR image and generates satisfying 
elevation data of the ground based on the shape-from-shape 
principle. Compared with other approaches in which two or more 
SAR acquisitions are absolutely needed, single image is 
required for DEM reconstruction in radarclinometry. Although 
there are constraints in the accuracy and some radiometric 
hypotheses such as the homogeneity of the surface and the limit 
of the slope, it is still promising in some geographic applications. 
Moreover, It is possible to integrate the radarclinometry with 
radargrammetric or interferometric approach to simplify the 
procedures of the data processing. Radarclinometry was 
developed by Horn and Wildey etc. (Toutin, 2000) and further 
investigation is still limited. 
In the following sections, the methodology of 
radarclinometry is briefly described. The elevation reconstruction