Automatic Recognition of Impact Craters on the Martian Surface from DEM and Images - Planetary Exploration and Science: Recent Results and Advances - page 110

methods use different algorithms to assign those parameters and in principle can be
even less correct. The issue is how to reconcile those measurements. In an attempt to
solve this issue, there is a framework proposed by Salamuniccar (Salamuniccar and
Loncaric 2008 ) for the solution of objective CDA evaluation which is an effective
approach for this problem.
The proposed framework consists of the following elements that must be strictly
defined for objective evaluation of CDAs:

1. Measurement of differences between craters
2. Specification of test field data
3. Specification of a ground truth catalogue
4. Selection of methodology for training and testing
5. Measurement of CDA performance

In order to produce subjective results, it is necessary to develop a solution to
automate the matching process. According to craters' radiuses, the relative values
represented the differences in position and size between two craters can be obtained.
A recommended definition is introduced in this paper for the measurement of crater
difference f m .Let r 1 and r 2 be radiuses of two craters, where r 1 r 2 ;let d be the
distance between the crater centers, and let f c 0 be a constant value called craters'
difference factor. If and only if the following two equations are satisfied, the two
craters are considered to belong to the same crater:
f m D max r 1

d
r 2

r 2 1;

(6.14)

f m <f c (6.15)

If the threshold is defined for correct matching f c , it can be determined by
checking all possible combinations within the GT catalogue.

6.4 Results and Discussion

In this chapter, we have already presented four comprehensive methods in automat-
ically detecting craters from images and from digital topography separately. Each
of them has shown their abilities in detection of craters. In order to evaluate the
performance of different methods, the quality factors true detection rate (TDR) and
false detection rate (FDR) are introduced, and they are computed as follows:

TD
GT 100 %

TDR D (6.16)

FD
TD C FD 100 %

FDR D (6.17)