a.l Regular grid: 
In order to use off-line Composite Sampling, 
terrain relief was represented by a dense regular 
grid, with a grid spacing of 25 m. 
a.2 Sampled information: 
The Z-information contained 1601 points in vector 
form, sampled selectively, which are arranged in 
tree subsets, namely; 
L-set 1 
I-set 2 
I-set 3 
The terrain relief was represented by 16000 points 
which are arranged in a regular grid forming 250 
(25x10) patches of 8x8 points each. 
a.3 Preprocessing 
a.3.1 E-sets 
I-sets were mapped into the grid domain. The 
rasterised format I-sets were then segmented into 
44 overlapping patches of 33%33 points each. 
a.3.2 Regular grid 
The regular grid consisting of 250 patches (of 8x8 
points each) was segmented into 44 overlapping 
patches of 33x33 points each. 
b. Progressive and composite sampling, TEST 1 : PS 
Input: 
- Regular grid DTM consisting of 44 overlapping 
patches of 33x33 points each. 
- Threshold (for second differences) 
Th - 10.0 m (S - 1/16) 
Maximum height difference AH in the test 
  
area = 155.214 m max 
| zest CASE| S = | VARIANT | G |HATER | E 4 
  
TEST 1 | 1/16 | PS |2.0 % [12.0 % | 35 % 
Table 2: Performance measures for PS 
TEST 2 : VARIANT-1; CS(1) 
Input: 
- Same as previous test +I-set 1 
TEST case | S 
| TEST 2 
  
  
  
     
      
Table 3: Performance measures for CS(1) 
TEST 3: VARIANT-2; CS(2) 
Input: 
- Same as previous test +Z-set 2 
  
  
  
  
  
  
TEST CASE| S = |vaRTANT | = [MAKER | E 
| TEST 3 | 1/16 | CS(2) [1.9 % 5.0 % 29 7 
  
Table 4: Performance measures for CS(2) 
TEST 4: VARIANT-3; CS(3) 
Input: 
- Same as previous test +Z-set 3 
TEST CASE| $ = | VARIANT 
ES 4 | 1/16 | CS(3) 
Table 5: Performance measures for CS(3) 
  
   
C. Performance estimates for progressive and 
composite sampling 
84 
  
  
  
PATCH |VARIANT| © c |MAXER | PS 
% OF Z PTS 
Li [CS(3), [0.832 (1.8 % | 8.30 [3434 
to (CS(2) {0.86 |1.9 z | 8.30 |3404 
11,4 |CS(1) |0.89 |2.0 x | 8.30 |3733 
PS 0.91 [2.0 x |19.65 |3997 
  
  
  
  
  
  
Table 6: performance estimates for progressive and 
composite sampling 
CS(3) = I-set contains all the break lines vhich 
fulfil the requirement of rule base. 
CS(2) = E-set contains all the previous break 
lines, except the break lines joining the 
peaks. 
CS(1) = E-set contains only Main break lines vhich 
are at the same time peripheral lines of 
anomalous regions. 
In order to reflect the role of £ information in 
the sampling, results of different tests were 
compared. 
TEST 
CS(3) 
versus 1/16 
PS 
CS(2) 
versus 
PS 
  
S R 
o 
R 
  
  
  
  
max 
  
    
   
| 11 | 2.37 
1/16 | 06 | 2.37 
  
Table 7: Performance estimation of different 
variants of CS with respect to PS 
j- 
1. 
CS(1) 
1 
versus 1/16 
PS 
.02 2.37 
  
  
  
In conclusion we can state, by using the break 
lines and break points which fulfil the 
specifications of the rule base in CS(3), .that 
apart from a grate improvement in the accuracy of 
the skeleton information, the overall accuracy and 
overall efficiency are also improved 
significantly, compared to PS 
(Rg = 11% and Rp = 17% ). 
When omitting the peaks and the auxiliary lines 
joining these peaks, the gain in overall accuracy 
is reduced by 5X and overall efficiency did not 
changed (compared to CS(3)). 
When omitting the break lines and break points 
which fulfil the specifications of the rule base, 
and by using only the main break. lines, the gain 
in the overall accuracy is reduced by 9% and the 
gain in overall efficiency is improved by 11% 
(compared to CS(3)). 
  
Fig. 10a: Contour map, Haifa 
4.1.2 Bonnieux region This model is partly 
covered by flat, and partly by accidental terrain. 
This justifies perfectly the use of optimum 
sampling Aerial photos were 23 + 23 cm, Scale - 
1/15.000, c - 150 mm Camera type = Wild RC 10, the 
   
Easti 
841.€ 
the e 
m ar 
DSR-] 
  
a. Se 
delin 
the 
peri 
selec 
Morec 
break 
the 
E-inf 
3.1 R 
In o 
sampl 
dense 
4.2 S 
The 
form, 
Ihe t 
which 
(16x8 
a.3.1 
grid 
patch 
8.3.2