The geology.at feature Prolog primitive (Appendix 1) can 
access raster data which represents a geological map. 
The pixel values returned by this primitive correspond to 
different geological units. The geological_unit fact acts as 
a key linking this unit number with a high level description 
of the geology e.g. 
geological_unit(11, ‘permanent ice’). 
geological_unit(51, knobby hummocky terrain’). 
There are also some facts which are not directly usable 
by the predicates in MIS, but which have been included so 
that various constants and physical values for Mars can 
be looked up without consulting a textbook e.g. 
atmosphere (‘carbon dioxide’, 95.32, percent,carr). 
When performing stereo matching using images selected 
by MIS, a record is kept of all the images which produced 
successful and unsuccessful stereo matches. This infor- 
mation is stored in example_stereo_pair facts so that future 
searches for stereo pairs could avoid unnecessary calcu- 
lation e.g. 
example.stereo.pair('635A83', '603A41', good). 
example.stereo pair('323S41', '639A92', bad). 
It is often necessary to find images which illustrate a par- 
ticular type of feature. Descriptions of all the Viking Or- 
biter images are not generally available in digital form, al- 
though catalogues of images containing clouds (Jansma, 
1992) and image clarity (Bolef et al, 1982) do exist. To 
solve this problem, two sources of example descriptive 
information can be used and entered as lists in Prolog 
facts. The first source is from photographic illustrations 
and captions contained in geological books and papers 
about Mars. The second source is from visual interpreta- 
tion of images which are recorded whenever members of 
the EXODUS project view an image in the course of their 
work. A reference key, indicating the source of the in- 
formation is included with these facts. Some of the words 
used in the image descriptions may be unfamiliar to users, 
therefore "dictionary" facts are included where explana- 
tions are needed. A third potential source might be image 
understanding routines which scan images or DEMs for 
particular features. 
example jmage('041A11', [volcano, caldera, 'orographic 
clouds, carr). 
example jmage('437S03' [| boomerang shaped structures! 
cook). 
dictionary('orographic clouds', 'clouds formed during uplift 
of air over large scale features). 
Books and scientific papers are also a useful source of 
facts about estimated crater retention age of features 
on Mars. In the example below, the typical, minimum, 
and maximum estimated ages (thousand million years) of 
Olympus Mons are given (Carr, 1981). 
example-crater-age(’Olympus Mons’, 0.1, 0.1, 0.3, billion 
years’, carr). 
Finally, some of the Prolog predicates to be discussed 
790 
in section 3.1, on stereo image pair selection, rely upon 
constant values or limits. These can be expressed as 
“rule” facts. In the examples below, which refer to the 
selection of stereo image pairs, the first argument acts as 
a rule identifier. The second argument is acted upon by 
the predicate using this fact. The final argument is used 
either as a comment statement or as a unit specifier e.g. 
rule('maximum stereo precision value’, 1000, metres). 
rule('maximum difference in solar altitude’, 5, degrees). 
rule(’'maximum difference in solar azimuth’, 45, degrees). 
rule(’'maximum stereo illumination difference’, 10, x10). 
rule(’'maximum stereo IFOV’, 650, metres). 
rule(’'minimum overlap proportion’, 0.05, fraction). 
These particular values were tightly chosen so as to im- 
prove the chances of the selected image pairs matching, 
and a height precision of better than 1km. Such tight cri- 
teria may result in a smaller number of stereo pairs be- 
ing selected and consequently poor spatial coverage (See 
Figure 1). To improve spatial coverage, "rule" fact values 
can be relaxed, however this is usually at the expense 
of the proportion of the images which can be matched or 
with a lower height precision. 
2.3 Prolog predicates 
Prolog predicates are assertions of facts or rules, which 
may be expressed in terms of other predicates (Clocksin 
and Mellish, 1981). This is illustrated in the moder- 
ate_size_feature predicate described below: 
moderate.size feature(Name):— 
feature_size(Name, Size), 
between(10, Size, 50). 
feature_size(Name, Size):— 
feature(Name, Lon, Lat, Size). 
feature_size(Name, Size):— 
feature(Name, Lat1,Lat2,Lon1,Lon2, Size). 
The purpose of this predicate is to determine whether a 
cartographic feature can be given an arbitrary class of 
“moderate-size”. A feature is defined as “moderate” if it 
is between 10 and 50km in diameter or length. A second 
predicate, feature_size is used to return the size of a fea- 
ture. This is necessary because there are two classes of 
feature, circular and areal/linear, and the latter has lon- 
gitude and latitude ranges. A selection of some of the 
standard predicates used by MIS is given in Appendix 2. 
3. APPLICATIONS 
3.1 Selection of stereo pairs 
Experiments undertaken during the EXODUS project have 
shown that the selection of candidate stereo image pairs, 
solely based upon image overlaps and having a suffi- 
ciently good base to height ratio, does not necessarily 
guarantee that they will successfully automatically stereo 
match. Early results suggested that even though the 
stereo matcher uses radiometric normalization, the BRDF 
of the surface changes the appearance of features dra- 
matically (Muller et al, 1992). Other factors include 
pr 
gh 
lis