004
; are
how
ocal
mal
(4)
ified
these
1S:
(5)
local
relief
at the
large
of the
ziness
fuzzy
elis
of the
sed on
pe, a
soning
n both
of the
reDiff)
sed to
ze. By
of the
| if it
livided
tisfied,
t The
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
linguistic variables, labels and the corresponding membership
functions for the fuzzy region growing process are given in
Table 1.
As an example, suppose that a building with its nearby tree in
the object space is extracted and classified as a single 3D
object. This inevitable miss classification, is revised by the
fuzzy based region growing in the image space. After the fuzzy
based consistency check, the region is subdivided into two
uniformly varying sections and subsequently will be treated as
two different objects.
2.2 Recognition
As mentioned above, our recognition strategy is based on the
concept of information fusion. These descriptors are important
elements for a comprehensive recognition process and they
need to be fed simultaneously into the recognition engine
(Samadzadegan, 2002). In order to make it quite clear, it should
be emphasized that in the previous stage the structural
parameter was used to determine only the presence and the
location of the objects. In the recognition stage, however, the
structural parameters are used for the object recognition
process. Thus, the extracted parameters are regarded as the
signatures expressing the object’s identity.
The structural descriptors of a 3D object are: height, area,
shape and relief variations. The shape of an object is expressed
by the length to the width ratio. To express the relief variations
for a 3D object, a relief descriptor is determined using the
indicator k expressed by Equation 5. The object’s structural
descriptors are indeed an efficient mechanism by which a
reliable recognition of many 3D objects can be conducted
without further involvement with the textural complications.
As mentioned earlier, the object recognition potentials can be
enhanced by a simultaneous fusion of the extracted STS
parameters. However, these descriptors are not crisp in their
nature and hence can not be realistically described by a rigorous
mathematical model. Therefore, our proposed method again
takes advantage of the fuzzy logic concepts to describe the
objects more realistically and consequently to perform objects
recognition process based on the fuzzy decision making
approach.
It is important to mention that, in principle, the descriptors are
not necessarily limited to the STS values. That is, the
information fusion process may also include other types of
descriptors if they are available. If, on the other hand, there are
only one or two descriptors (e.g. only spectral, or spectral and
structural), the recognition process can still be executed, but
this time, of course, giving rise to a less reliable result. The
first step in proposed recognition process is to determine the
degree to which the STS descriptors belong to each of the
appropriate fuzzy sets via membership functions. The linguistic
variables which are used for each of structural, textural and
spectral descriptors are presented in Table 2.
Once the STS components have been fuzzified, the sequential
fuzzy reasoning procedures, comprising: Implication,
aggregation, and deffuzification, are performed (see Section
371.2).
3. EVALUATION OF THE PROPOSED OER
STRATEGY
To assess the capabilities of the proposed OER method a
sample LIDAR data of an urban area of city of Castrop-Rauxel
which is located in the west of Germany, was selected (Figure
3). The selected area was suitable for the evaluation of the
proposed OER method because the required complexities (e.g.
proximities of different objects: building and tree) were
available in the image (Figure 4).
/2
4
ar
il
x
RR
TAN
Before the system operation is started it is necessary to set up
the fuzzy reasoning parameters. For the BT object classes, the
preliminary membership functions for the ST components are
defined based on the knowledge of an experienced
photogrammetric operator (Figure 5 and Figure 6).
3.1 Operational stages
The OER process was initiated with fuzzy based region
extractions operation and hence effectively regions are
constructed (Figure 5). In the next step the extracted regions are
analyzed to derive the ST descriptors. In the next stage, the
recognition operation is activated by which all BT objects in the
sampled area patch were successfully recognized. Figure 7.
shows the output of each stage for a sample image patch and all
recognized BT objects for the entire test area.
Table 2. Linguistic variables and labels of fuzzy reasoning structure in recognition process
Type Linguistic Variable Linguistic Labels
Height SoShort ,Short, Medium , High , SoHigh
Structural Area SoSmall , Small , Medium , Large , SoLarge
Relife Solrregular, Irregular , Regular , SoRegular
Input Shape NonStretched , Stretched , soStretched
Textural Texture Solrregular, Irregular, Regular , SoRegular
Output | Object Object Type Not , ProbablyNot , ProbablyYes , Yes
923
