both 
n on 
dges 
ps to 
ician 
ause 
hese 
steps 
d to 
cting 
ent. 
atical 
erent 
n as 
lines 
inary 
als a 
scale 
sent 
f the 
mbra 
inion 
ence 
gy is 
place 
  
GSK(X) = XO{L} (3) 
where X refers to contour map image, “O" represents 
gray-scale thinning, result in a gray-scale thinned contour 
image. Because of the finite width of scanned contour 
(often several pixel width),the operation convergences 
quickly after no more than 2 or three recursive times. The 
result preserves the contour in middle axis, disappear the 
submerge mostly. Furthermore, the result opened with 
the structure element H clears the submerge completely. 
That is 
GSK'(X) 2 GSK(X)o H (4) 
1 1 1 
where "o" represents open operation, H=1 (D l1. 
]. 1 i 
Moreover, the contour now is sometimes in jagged way, 
and its structure reflects as 
0 
0 
m— — C 
1 
0 
; mc ie 
i7 1,2,...8 rotate 1 as soon as i increase 1 
Design these group feature as structuring elements ( J; ) , 
by means of thinning transformation as 
GSK"(.X) 2 GSK'(X)O (J;) (5) 
leads to one pixel width contour. 
The gray tone contour image now reveals a surprising 
by-product, that is, the gray level histogram concentrates 
on the maximum gray value. So luckily, it is easy to 
determine threshold to segment the contour from the 
image, The difficult segment problem is solved! 
4. AUTOMATIC RASTER TO 
VECTOR CONVERSION 
4.1 Data Structure 
The linear quadtree is a data structure which is based on 
the regular decomposition of a square into quadrants and 
sub-quadrants[6]. It represents spatial relations only by 
leaf nodes with numeric keys. Contour map after 
thresholding is organized in such structure not only 
compact raster data in a high rate, but also improve the 
operation speed on morphological transform. 
Freeman code is often used to describe a vector data in 
the form of number series. The number representing the 
dot near the center is less than 9 when the data are 
organized in 8-connect form, such as 
top 
4 3 2 
left 5 0 1 right 
6 7 8 
bottom 
Using this code to describe a contour helps to learn the 
Properties of a contour such as length, direction and 
Connectivity. Obviously these features are important in 
contour line identification. The dot on a contour can be 
531 
caught by recursive morphological thinning with structure 
elements { E;) such as 
eui 0 
ooo 
CS — © 
i=1,2,...8 rotate T as soon as i increase 1 
and the Freeman code of the dot is equal to the number i 
of series ( E; ).. 
4.2 Contour Line Extraction 
To extract contour line from annotation symbols on a 
contour map is the first step of raster to vector 
conversion. In general , the length of contour is far from 
the length of symbols, and contour is not crossed by 
themselves or each other. Although this way can 
separate most contour from the other part of map, some 
little arcs which are broken from the contour by manual 
annotation are filtered also, but they are often important 
to show features of the contour. So the method should be 
improved. 
In order to provide a precise mathematical description of 
the important shape-size of a scanned contour map 
image under consideration, the pattern spectrum is 
utilized. The pattern spectrum PSK(X) of a set X in terms 
of the structuring element B is given by 
Card[Y, (X)-Y,,,(X)], fork >=0 6 
Card[O. , (X) - b , ,(X)],fork «- -1 (6) 
where k €Z and Card[X] denotes the cardinality of a set X, 
Y represents open operation and ® represents close 
operation. The pattern spectrum provides a measure of 
similarity between an image and the collection of all 
possible structure elements kB. So selecting suitable 
structure elements to describe contour line not annotation 
symbols such as 
PSK(X) = 
© © 
ooo 
— © © 
i 2 1,2,...8 rotate E as soon as i increase 1, and 
1.0.0 
0. 0 0 
0 1. 0 
i Tc 2 
i 7 1,2,...8 rotate 1 as soon as i increase 1 
contour line contributes to lower size part of the pattern 
spectrum, where annotation symbols contribute to higher 
size part of the pattern spectrum. Therefore the contour 
line is distinguished from annotation symbols and is 
extracted easily. As you may know, the method is also 
suitable for annotation symbols recognition. 
4.3 Contour Gap Connecting 
It is necessary to deal with the *bottle-neck" problem of 
contour gap connecting, in order to set topological 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996