IMAGE RESTORATION METHODS AS PREPROCESSING TOOLS IN DIGITAL STEREO 
MATCHING 
Jan-Olov Fallvik 
Department of Photogrammetry 
Royal Institute of Technology 
s-100 44 Stockholm, Sweden 
ABSTRACT 
In this paper basic restoration techniques for the purpose of in- 
creasing matching accuracy in digital stereophotos are investigated. 
Inverse filter, Wiener filter, constrained filter and geometric mean 
filter are discussed. These traditional restoration techniques are 
based on stationary random fields or spatial invariant point-spread- 
£unctions or both these criteria. The result shows that other methods 
based on the actual physical background are required, which means that 
these methods must be able to process nonstationary images and spatial 
variant point-spread-functions. 
1, INTRODUCTION 
During recent years much effort has been invested in digital methods in 
analytical photogrammetry with the focus on digital image processing 
methods, The goal of photogrammetry is to describe properties of 
objects, such as shape and position in an image. The extraction of 
shapes belongs to the field of pattern recognition and feature ex- 
traction. The determination of object position may be performed using 
digital matching techniques. 
An investigation on the geometric accuracy of the matching of ohjects 
in simulated SPOT stereo images has been performed at the Department of 
Photogrammetry at the Royal Institute of Technology, Stockholm 
(Rosenholm, 1985). He reports an expected accuracy o£ 0.1 pixels and 
experimental values of 0.3 pixels. The question is, if there is any 
possibility of increasing the matching accuracy by preprocessing the 
images in a suitable manner. The answer may be £ound by considering 
the problem as à restoration problem. If the undegraded image can be 
well recovered, the matching accuracy may increase. 
For the purpose of improving digitized aerial photography, à number of 
basic digital image restoration methods are described in this report. 
By image restoration we here mean the removal or reduction of de- 
gradations that where incurred during the imaging process. Typical de- 
gradation factors include camera motion, lens aberrations, film grain 
noise, low pass filtering due to the electro-optical systems and 
atmospheric turbulence. 
Earlier work in the area of preprocessing due to stereo image matching 
has been performed by Forstner (1982) and Ehlers (1982). Forstner 
derived a method of optimizing digital image correlation. By mini- 
mizing the variance of the estimated difference of image translation, 
Forstner designed an ideal low pass filter with an optimum cut-off 
- 223 7