area 
m the 
n the 
least 
TM 
4 can 
. In 
isplay 
r than 
ay has 
ation. 
can be 
res of 
are 
J the 
21pful 
rsheds 
image 
a 3x3 
38. If 
Lowing 
ridge: 
. when 
ring 8 
2. Height of pixel P should be the highest on 
one of four profiles which pass through 
pixel P. 
3. Pixel P is not a single pixel. 
  
  
  
  
  
  
  
  
Figure 9 A 3x3 operating mask 
The criterion 1 and 2 are opposite in extracting 
valleys. The shadow of terrain can be gained 
by hill shading method. The simulated sunshined 
images is produced by computing the density of 
reflection according to the difference of 
terrain releaf, incident light source and 
position of viewpoint. Therefore, before 
computation has been accomplished, the positions 
positions of sun light and viewpoint must be 
defined. The simulated sunshined images has 
geometry of ortho- projection and the appearance 
of terrain releaf is augmented by shadow to 
present a stereo effect [5]. This approach has 
been applied in computer assisted mapping and 
computer vision [18,2]. 
IV. THREE DIMENSIONAL IMAGES REPRESENTATION FOR 
COMPUTER VISION 
Terrain surface has the property of random 
releaf and contains a plenty of information of 
features. The images from remote sensors 
presents only planar scenery and lacks a stereo 
vision effect. Three dimensional images display 
has effect of stereo vision and is completed 
through a transformation of mathematic 
projection using both remote sensed images and 
DTM data. This approach is called landscape 
visualization [8,11] and has been applied to 
many disciplines such as GIS, war game and 
flight simulation etc. In computer graphics, 3D 
images display is performed by 3D transformation 
using a homogeneous coordinate system [17]. 
This approach is also applied to landscape 
visualiztion [7,13]. In this paper, we introduce 
the colinear equations which are well known in 
photogrammetry. The colinear equations are used 
to describe the relationship between object 
space and projective plane with computational 
vision method while the spatial position of 
viewpoint is given. This approach is convenient 
for the setup or simulation of conditional 
parameters of actual visualiztion. In this 
section, we will illustrate the data 
preprocessing and related algorithms design. 
4-1 THE PREPROCESSING OF IMAGES AND DTM DATA 
The purpose of preprocessing is to make each 
pixel on images coincidence with a ground 
elevation. Therefore, the procedures include 
increasing DTM density by interpolation and 
ortho-rectification of images. If the MSS 
images from satellite is used, then color 
enhancement must be performed first. This 
enhancement can make images better color 
effect. We can derive the true color scene 
from RGB bands of  LANDSAT TM images. However, 
of enough color 
although the 
it still presents the lacks 
satuation and causes lower hue 
905 
contrast enhancement has been done [4]. For 
overcoming this problem, the color coordinate 
transformation is used to achieve color 
enhancement [9]. 
4-2 TRANSFORMATION OF PARALLEL PROJECTION 
defined at infinite in 
Hence, the 3D coordintes 
space are sequently 
along the direction 
The viewpoint is 
parallel projection. 
of points in object 
projected on 2D plane 
parallel to the sight line. The object 
coordinates  (Xp,Yp,Zp) of object point P is 
transformed to the image coordinates (xq,yq) of 
projective point Q, i.e. 
Hg = Hp COS à + Vp Sin o 
Ug - Zp cos « (-Hpsino « Vpcosa-D sing (7) 
where is the azimuth of sight line, and is 
the roll of sight line. 
4-3 TRANSFORMATION OF PERSPECTIVE PROJECTION 
If the orientation parameters 
center (viewpoint) and the location of 
projective plane are decided, then the 3D 
coordinates of object can be projected on 2D 
plane. The formulas are well known in 
photogrammetry and written as: 
saul - Hg) * aj2(V - Ya) + a13(Zp - Zg) 
asi(Hp - Hg) + as2(¥p - Ye) + a33(Zp - Za) 
zf ax (Hp = Ha) + 322(V, - Vg) + a23(Zp = Zg) (8) 
asi(Hp - Ha) ^ as2(Vp - Vo) * ass(Zp - Ze) 
of perspective 
  
Hg = 
  
apy = COS f COS o 
312 - COS f Sin a 
313 - - Sin f 
321 - Sin y Sin cosa - CcOSy Sino 
322 - Sin y Sin f sin a.» COSy COS o (9) 
325 - Sin y cos f 
331 - COS y Sin f cosa - Siny Sino 
432 - COS y Sin f sina. - Sin y COS o 
333 - COS y COS f 
where o is the yaw of sight line, 
Bis the pitch of sight line, and 
y is the roll of sight line. 
4-4 TRANSFORMATION OF PANORAMIC PROJECTION 
Panoramic projection is similar as cylinder 
projection and the viewpoint O is located at 
the central axis of cylinder. Then the object 
point P is project on the plane of cylinder and 
the equations are written as: 
Tq = ap r 10 
r 
Zo t j- (Zp 7 Zo) 
r 
Il 
Yq 
where a is the polar angle of object point P, 
1 is the distance between object point 
and central axis of cylinder, 
z is the coordinate of cylinder 
and r is the radius of cylinder. 
system, 
4-5 ALGORITHM DESIGN 
The purpose of projective transformation is to 
project the 3D coordinates on 2D display 
plane. In fact, because the terrain releaf is 
different in different place, there are many 
hidden points to be processed during the