International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
where TP: TRUE POSITIVE - Both data sets (detected crater 
and comparison data set) classify the pixel as being part of a 
crater 
TN: TRUE NEGATIVE - Both data sets classify the pixel as 
being part of the background 
FP: FALSE POSITIVE - Detected data set classifies the pixel 
as a crater, comparison data set classifies it as background. 
FN: FALSE NEGATIVE - Detected data set classifies the pixel 
as background, comparison data set classifies it as a crater 
However, if we apply these factors directly to the optical image 
results, the detection ratio is likely to be very low, because the 
non-detection of large sized impact craters significantly reduces 
the detection ratio. So the QA factors are modified for impact 
craters, which have higher detection percentages than 50%, 
which are considered as true detection. Detection results by this 
scheme are shown in Table 1. 
   
   
    
(a) Crater detection on MOC- | (b) Crater detection on MOC- 
WA level 2 image WA level 2 image (M0203967) 
M0203967) 
  
   
  
  
e 
(48 E git + 
(e) Crater det (f) Crater detection on MOC- 
WA level 1 image M0300749 | WA level 2 image (M0900109) 
Figure 9. Crater detection examples for craters of various sizes 
and shapes. Note that heavily eroded craters are not 
currently detected, 
   
  
  
  
  
3 
(a) Centre points of craters in (b) Crater detection results 
MCC overlaid on MOLA with MOC-WA level 2 image 
DEM (M0101958) same area with 
case (b) 
  
  
  
  
Figure 10. Inter-comparison with MCC data sets and detected 
craters on MOC WA image (white + ; Barlow data 
sets , Black + : Kuzmin data sets) 
  
  
  
  
  
  
  
  
  
Manual measurement 
MCC Small 
; Large 
(Barlow) size n 
(R «8 ne 
v (8 <R<60) 
pixels ) 
True 
positively 
detected 60 198 120 
crater 
number 
False False positive has no meaning 
Dositively for the MCC inter 
: comparison, because MCC 
detected Smpaîrss Sod s 74 10 
data sets don't aim to 
crater : 
catalogue all craters in target 
number 
area 
False 
negatively 
detected 12 32 12 
crater 
number 
Detection ; 
83% 86% 90% 
percentage 
Branching 
Homns - 0.37 0.08 
factor 
ualit 
Quality j 6596 84% 
percentage 
  
  
  
  
  
*True negative has no meaning for individual crater detection 
Table 1. Impact crater detection ratio by intercomparison with 
MCC and manual measurement in 12 random MOC 
WA images 
3.3 Simulation with different illumination condition 
As far as we have experienced, the illumination condition is 
crucial for the positional accuracy of the detected impact crater. 
To assess the robustness of an algorithm with different ranges 
of illumination angles, simulated crater DTMs whose diameter 
was 100 pixels and which have vertical 3D profiles of well 
known craters proposed by Duxbury (1991), was employed. 
Hill shaded images were generated at 10° intervals within the 
range of 0-360° sun azimuth angle and 20-90° elevation angles 
using a Minnaert surface’s reflectance model which is given by 
(8). 
P = B(cosi )' (cose )'" (8) 
where B: brightness coefficient , 
k : the constant of Minnaert exponent 
e: emission angle, I ; incidence angle 
The results show very good agreement with actual (modelled) 
positions except at very high sun elevation angles around 90°, 
which rarely occurs for real image acquisitions. 
820 
Intert 
  
Figur 
Autor 
to id 
condi 
comp 
detec 
condi 
again 
Curre 
diame 
great 
a mu 
releas 
HRS( 
practi 
Ather 
detec 
Barlo 
impac 
Pasad 
http:/, 
Burl, 
1999, 
Minir 
and K 
Yuen 
Partia 
Image