lines on gray tone image and disappear background 
optical illumination submerge. Luckily, gray-scale 
thinning, an efficient algorithm is found to deal with it 
powerfully. Thirdly, contour lines in raster form must be 
changed into vector form to construct a topological 
database for terrain analysis. It is necessary to set up 
topological relationship in spite of poor broken contour 
segments. That’s why extracting contour lines from 
annotation symbols, gap connecting be studied. 
Furthermore, contour elevation must be determined and 
a new way by means of geomophological points and 
lines is presented. Finally, corresponding digital elevation 
model (DEM) is constructed in the form of triangulated 
irregular networks(TIN). 
3. SCANNED CONTOUR MAP IMAGE PROCESSING 
3.1. Data Capture 
There are two reasons about contour map scanned in 
gray tone image rather than binary image: one is about 
the scanner, the other is about map itself. The point 
spread function of the scanner leads image to 
convolution distortion and nonuniform illumination with 
nonuniform paper reflection make map background 
submerge. On the other hand, different types of pens and 
ink might be used on the map to distinguish various kind 
of symbols. 
3.2. Rectification And Resample 
Map image from scanner has two aspects deform: one is 
on scanner, the other is on map itself. As for the former, 
a standard lattice is used to test rectification coefficients 
in the form of quadratic polynomial: 
x, =a,+ax+ay+ax*x+a,x*y+a;y*y 1 
yy 7 by tbx t by t bx * x t bax* y c bsy* y m 
As for the latter, the kilometer lattice on the map can be 
used to measure the extend of deform in the form of 
quadratic polynomial also: 
X, 76 tox toytox*xtex*ytoy*y 
y,7dy*da*d,y*dy*x*dy*ysdy*y — 
After parameter ai,bji,ci,dj can be solved by the least 
squares method, each pixel on the image corresponds to 
a new position(x2,y2). Because x2,y2 Maybe not integral, 
so it must give a correct gray value to each pixel ‚that is 
image resample. The resample method is selected as 
bilinear interpolation[1]. 
3.3. Image Filter 
Image filter is dealt with in both frequency domain and 
spatial domain. In frequency domain, Fourier analysis 
method is used to improve image quality because it 
corresponds to optical scanner characteristic. In order to 
preserve contour, it must remain the high frequency 
component of the image, and restrain the low frequency 
component of the image. So the homomorphic filter is 
selected and the coefficients of both high frequency and 
low frequency are carefully chosen. 
Recently, wavelet transform is widely used in image 
processing. Superior to Fourier analysis, wavelet 
transform contributes good localization properties in both 
the spatial and frequency domain. Wavelet transform on 
the scanned image enhances noise-corrupted edges 
under the multiresolution decomposition and this helps to 
detect contour lines. 
Traditional image processing methods such as Laplacian 
enhancement and midvalue filter are not used because 
they haven't selectivity of spatial localization. These 
methods smooth the contour and make followed steps 
more difficult. Instead, morphological filter is used to 
delete little noise such as grain and short arcs. Selecting 
3*3 structuring elements as follows: 
0370,70 
lO d 
0 0 0 
c 
i=1,2,...8, rotate y as soon as i increase 1, and 
0.0 0 
0-Q 1 
g.1. 0 
a 
i=1,2,...8, rotate > as soon as i increase 1 
where "(D" represents the centre of structure element. 
The recursive opening operations of mathematical 
morphology [3] are needed to eliminate noise in different 
size and conditional dilation operations are chosen as 
assisting tool to control the lose of contour lines 
information. 
3.4. Edge Detection 
Gray-scale morphology is natural extend of binary 
morphology from two to three dimensions, it reveals a 
surprising landscape of beauty and utility[4]. Gray-scale 
morphology uses the basic concept umbra to represent 
grayscale data. The significance of the umbra of the 
umbra of image processing is that they remain umbra 
under the usual morphological transformations of union 
and intersection, dilation and erosion. The difference 
between grayscale morphology and binary morphology is 
only that the operations in the first domain replace 
intersection and union with min and max. 
Select flat structure elements {Lj} as 
0.0; 0 
spo 
iad 1 
c 
i=1,3,5,7, rotate > as soon as i increase 2, and 
21:0 0 
10 0 
1 
c 
i7 2, 4, 6, 8, rotate 2 as soon as i increase 2 
where *"." represents O or 1. 
Operate 
530 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996 
  
wh 
gre 
im: 
(of 
qui 
res 
sul 
the 
Th: 
wh 
Mo 
anc 
Des 
by | 
leac 
The 
by-} 
on 
dete 
ima 
4.1 
The 
the 
sub: 
leaf 
thre 
com 
opel 
Free 
the 
dot 
orga 
Usin 
prop 
conr 
cont