18
as
ke
to
CS
of
nt
W
of
Or
of
1S
of
and stored together with the label in a database. Expert
system technique is introduced during the detection of
buildings with these descriptor values. After rules are
generated and tested and accuracy of interpretation
evaluated, ground coordinates of features of detected
buildings are computed and projected to an existing Geo-
Spatial Information System vector map.
2. THE DETECTION PROCESS
The detection process to automate the identification of
building changes is designed as follows.
Scan Aerial Photos Design
(Stereo Pair) Descriptors
: Using data
i from
; Existing
Image Segmentation Geo-Spatial
i Database
and
Extract Corner Extracted
Points Points
| |
Transform Extracted Form
Points to Ground Rulebase
Coordinates
Identify Buildings Produce Digital
with Point Symbols Orthophotos
Overlay and
Display Result
Aerial photos taken at a scale of about 1/3,000 are used
as base material for interpretation. The films are then
scanned and image segmentation is carried out with the
scanned images to segment into regions and also labeled.
Coordinates of corner points of segmented regions are
transformed to object space. To test if the labeled regions
are buildings, expert system is introduced. Descriptors of
989
regions are formulated to be used in the generation of a
rulebase which will decide if the selected region is a
building or not. A symbol will be generated at the
centroid of the region and placed to represent the
identified buildings.
The symbols of detected buildings is again superimposed
over the scene so that the user can visually confirm the
result of the identified building. Buildings vector
polygons with a symbol inside the polygon, will be
buildings that are unchanged between the production of
the vector maps and the exposure of the aerial photos.
Newly built buildings will be those symbols that do not
have an enclosing building vector polygon, and building
vector polygons without a symbol inside will those that
have been removed. Digital orthophotos are produced
and displayed onto the screen. Vector maps of the same
area, produced prior to the exposure of the aerial
photographs, are superimposed over digital orthophotos
as well as the point symbols representing detected
buildings.
3. APPLICATION OF IMAGE PROCESSING
Image segmentation is the process of generating uniform
and homogeneous areas for feature extraction. There are
many different methods to segment digital images such
as local line detection method, edge detection method,
region growing method. Although many researches are
being made to improve the quality of segmentation,
many methods are of the ad hoc type leaving the user
with no information about the quality of the result and
the segmentation is not performed under the restriction
of an object model requirements (Stokes, 1992.). Region
growing method such as blob coloring is commonly used.
The algorithm is described as using an L-shaped window
mask to scan an image (Ballard and Brown, 1982.). The
rows are examined by rows to check for differences in
gray levels and if the difference is beyond a set value
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996