Environmental Engineering Reference
In-Depth Information
of OBIA that splits an image into separated regions or objects
(Myint et al ., 2008). Lizarazo and Barros (2010) proposed a fuzzy-
based image segmentation procedure that combines properties
of fuzzy and crisp image regions for urban land-cover classes.
Walker and Blaschke (2008) employed a two-tier segmentation
procedure that originally segmented the entire scene with the
same parameterization for all objects and re-segmented based
on spectral heterogeneity of neighboring urban objects. Even
though there are many different segmentation algorithms exist
(Pal and Pal 1993, Blaschke, Burnett and Pekkarinen, 2004),
there are only very few operational commercial software package
available that support image segmentation and object-based clas-
sification. Currently, the majority of applications are built within
the software environment found in Definiens (Benz et al . 2004;
Walker and Blaschke, 2008). This software utilizes multiple seg-
mentations based on a global heterogeneity criterion (Baatz and
Schape, 2000). The segmentation function in Definiens software
(Baatz and Schape, 1999 and 2000) is based on three param-
eters, namely shape ( S sh ), compactness ( S cm ), and scale ( S sc )
parameters.
9.2.1.2 Object scale levels
Segmented objects are organized into image object levels also
known as scale levels. An image object level serves as an inter-
nal working area for the object-based image analysis. The scale
parameter that controls the object size that matches the user's
required level of detail is considered the most crucial parameter
of image segmentation. Different levels of object sizes can be
determined by applying different numbers in the scale function.
The higher number of scale (e.g., 50) generates larger homoge-
neous objects (smaller scale - lower level of detail), whereas the
smaller number of scale (e.g., 10) will lead to smaller objects
(larger scale). A smaller number used in the scale parameter
is considered higher level in the segmentation procedure. The
decision on the appropriate level of scale depends on the size of
objects required to achieve the mapping goal(s). The software
also allows users to assign different level of weights to different
bands in the selected image during image segmentation. This type
of scale is a spatially aggregated scale (few pixels vs. more pixels in
objects). Each image object level consists of a layer of segmented
objects. It is important to note that image layers are the data
that exist in the original image when it is imported. However,
object levels store image objects that represent different types
of image data. Figure 9.1 demonstrates how hierarchical image
segmentation delineates image objects at different levels.
An entire image can be segmented based on the three key
parameters (shape, compactness, and scale), with parameter
selection depending on the size and shape characteristics of
objects associated with the classes of interest. At this stage, we do
not know which objects belong to which class, but still need to
assign predetermined classes to segmented objects.
9.2.1.1 Compactness and shape factors
Users are required to specify values ranging from 0 to 1 for the
shape and compactness factors to determine objects at different
level of scales of image objects. These two parameters control
the homogeneity of different objects. The shape factor adjusts
spectral homogeneity vs. shape of objects. A lower value of
the shape parameter leads to lesser influence of color on the
segmentation. The compactness factor, balancing compactness
and smoothness, determines the object shape between smooth
boundaries and compact edges. The more compact image objects
can be achieved by specifying a higher compactness value. The
criteria for optimizing objects in relation to compactness is
used when delineating image objects that are compact, and are
separated from non-compact objects, but contain relatively weak
spectral contrast (Definiens, 2007).
9.2.2 Features
InDefiniens (or eCognition) software, a feature is an attribute that
represents certain information concerning segmented objects in
Object Scale Level 3
(Biggest Objects)
Object Scale Level 2
Object Scale Level 1
(Smallest Objects)
Original Image
(Pixel Level)
FIGURE 9.1 Image objects at each image object level.
Search WWH ::




Custom Search