(6) 
, C or B and 
> of type A, R, 
of type “roof” 
(7) 
moving of the 
(8) 
neration of the 
) duced starting 
(9) 
    
There is a peculiarity in application of the above rules. During 
the arc head movement a change of the arc type could happen, 
depending on the sequence of isolated arcs. The finishing rules 
for producing the right part of the wall arcs depend on the 
relative position of the adjacent walls (on same building, on the 
near or far from the projection centre. 
Jco Vx 2 EmJ ev Vx 
Joa Vx = Wide Vx 
JeaVx =» Wiz Ucy 
Jcalx > En Jey Tx 
Jcaïx — WiJeyJx (10) 
JcoJ x > Wis Ucy Tx 
JecaUx > ErJewLx 
JcaUx > WsJeyUx 
JeaUx > Ws Ucy Lx 
The similar form have rules for generation the left part of the 
wall border when the lower part of the wall is hidden by the 
part of the staying in front building. The starting rules have 
similar form if the wall is hidden by wall or by the roof. The 
more usual situation when the whole walls are visible tends to 
another set of production rules. 
Another set of rules is formulated for generation of arcs, 
corresponding to the projection of the roof over the wall. Three 
groups of rules could be mentioned for generation of roof / wall 
border: for arcs in the middle of the wall (type N), left side arcs 
(type F) and right side arcs (type G). 
A special groups of transition rules are connected with 
conversion from external to the internal contours. To ensure 
this process two stage procedure is used. At the first stage the 
scanning of all the isolated arcs from the new formed contour is 
applied. The rules use state modifier of type ®. For every 
contour is applied a procedure for internal contour scanning. 
After internal contour scanning it is possible to apply two type 
of production rules. - the rules for transition to terminal 
elements or the rules for generating of invisible lines. The 
procedure of generation of invisible lines starts from the main 
contour of the external level and moves around the all arc to 
convert them in status of generating the hidden lines. Such 
conversion is possible only for element of type "border" 
(excluding e and h arcs). During the conversion process the 
generating procedure finds the lowest level contour in the 
processed cluster. For that contour is generated contour of arcs 
of type t or hidden arcs from the border type arcs. The 
concurrent description of contours is produced that is similar to 
the description for the visible one but without elements of type 
"edge" (h, e). If contours from lower level are not attached to 
the upper level by element of type r, a change of the main 
element of the contour is produced. After that all arc elements 
of internal contour are converted to terminal state. The scanning 
proceed with the contours of the upper level until the highest 
level for current cluster is reached. During that process the 
description of internal contours is produced for which some 
arcs of type r or v could be included several times in the 
contours of different levels. The process of image description 
generation for simple block of three buildings with different 
height and disposition is shown in Fig.l. The corresponding 
producing rules are shown in the right part of the Figure 1. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
4. MAIN STEPS OF PROCESSING 
The first step of processing includes contour finding procedure. 
To ensure reliable closing of the contours it is appropriate to 
apply edge detector procedure after image segmentation step. In 
that case the contour detector could be applied only to that arcs 
of the image where preliminary area boundaries pass. 
The second step applies contour matching between two stereo 
images. To ensure good height accuracy a method for arc chain 
coding with subpixel resolution have to be used. This procedure 
is applied only to those arcs that are visible in two images if 
correspondence could be estimated. The estimated height 
properties of the arcs are attached to the generated arcs. 
The third step of processing is application of producing rules of 
the above defined grammar to connection of arcs with their 
height properties. The goal is to produce an image description 
that is allowed in terms of the defined grammar. The last stage 
of this step is the creation of a separate object description. 
5. DISCUSSIONS 
The suggested approach of description of images of urban areas 
could be applied after procedure of automatic contour 
extraction and contour matching in digital stereo images or after 
collection of vector data by a digital or an analytical plotter. 
The generation procedure could be applied for creation of a 
skeleton model of the object or for verification the accuracy of 
the arc labelling 
The problems that arises are that it is not possible to ensure 
height information for all visible arcs in two images. The 
second problem that arises is related with contour disconnection 
due to the presence of noise in input image and unperfected 
contour tracing procedure. As a result gaps in the model appear. 
To solve that problems an improvement of the grammar by 
adding of stochastic rules have to be done. 
Another problem is due to the fact that the suggested approach 
is based on height information of the arcs and especially the 
relation of arc heights in joint points. For such points a contour 
matching procedure could failed and to avoid this the height 
information have to be determined in the nearest to the 
connection point area of the image that produces some 
limitations for low resolution images or relatively small objects. 
The produced image model could be used for excluding the 
building areas from generation of DEM or for purposes of 
mosaicing of two stereo image during the process of digital 
photo transformation. For purposes of  orthophoto 
transformation a mosaic of the selected parts of the processed 
images could be created in which the hidden areas in one of the 
stereo images are taken from another image if corresponding 
area could be found in it. 
The filled up model containing information for the terrain and 
the artificial objects (buildings and etc.) could be applied also in 
a similar way for purposes of an improved land cover 
classification. The mosaicing of the areas of the two images 
could be done to complete the classified image of the terrain.