object-based change detection typically results in errors around real-world objects
such as buildings due to misregistration, different look angles, or difference in shadow
between dates (Im et al., 2008b; Stow, 2010).
Accuracy assessment of OBIA is another hot issue Clinton et al. (2010). Some
studies use point data to assess OBIA results, while others use polygons (i.e., objects)
as reference data. The same consideration of whether points or objects should be used
is also applicable to training data for supervised classification. While accuracy
assessment of OBIA is a somewhat controversial issue, there is a trend to develop
a standardized framework. Details about accuracy assessment of OBIA are found in
Radoux et al. (2010) and Lein (2012).
A fourth topic of consideration is optimum scale in OBIA. Image segmentation
algorithms influence the geometry of image objects, and thus it is crucial to
understand the parameters that are used in image segmentation algorithms and to
identify how those parameters affect image objects in terms of shape and size. A
successful segmentation would result in image objects similar to real-world
objects. When it comes to object size, most segmentation algorithms use some
type of scale parameter to control overall size of image objects. Many studies use a
multiscale approach to deal with the inherently different spectral and spatial
characteristics of disparate features such as buildings, roads, and forests within
a scene. Some studies have tried to identify optimum scales in OBIA. Selecting
optimum scale is often challenging as there is no standardized method to identify
the optimality; many studies simply adopt visual interpretation of objects or the
accuracy of subsequent analyses such as classification as selection criteria. In
addition, scales in most segmentation algorithms are arbitrary values and provide
only a relative comparison between the scales when the same segmentation
algorithm is used to the same input data. There should be standardized guidelines
regarding scales to facilitate the generalization of OBIA in remote sensing
applications, to enable efficient comparison of different OBIA approaches, and
to select optimum scales. The main objectives of this chapter are to provide a
review of recent publications (since 2008) in OBIA focusing on scale optimization
and discuss the related trends.
10.2 BRIEF OVERVIEW
We focused on peer-reviewed papers on OBIA published since 2008. The journals
we searched for papers include Remote Sensing of Environment (19), Photogrammetric Engineering and Remote Sensing (15), International Journal of Applied
Earth Observation and Geoinformation (8), ISPRS Journal of Photogrammetry and
Remote Sensing (8), IEEE Transactions on Geoscience and Remote Sensing (3),
International Journal of Remote Sensing (4), and others (19). We reviewed a total
of 76 papers, and Figures 10.1–10.3 provide a brief overview of the selected papers
for review. The number of publications has generally increased over the past five
years (Figure 10.1). Most of the selected studies utilized the commercial software
tool eCognition for performing the image segmentation that is a key step in OBIA
BRIEF OVERVIEW
199
such as buildings due to misregistration, different look angles, or difference in shadow
between dates (Im et al., 2008b; Stow, 2010).
Accuracy assessment of OBIA is another hot issue Clinton et al. (2010). Some
studies use point data to assess OBIA results, while others use polygons (i.e., objects)
as reference data. The same consideration of whether points or objects should be used
is also applicable to training data for supervised classification. While accuracy
assessment of OBIA is a somewhat controversial issue, there is a trend to develop
a standardized framework. Details about accuracy assessment of OBIA are found in
Radoux et al. (2010) and Lein (2012).
A fourth topic of consideration is optimum scale in OBIA. Image segmentation
algorithms influence the geometry of image objects, and thus it is crucial to
understand the parameters that are used in image segmentation algorithms and to
identify how those parameters affect image objects in terms of shape and size. A
successful segmentation would result in image objects similar to real-world
objects. When it comes to object size, most segmentation algorithms use some
type of scale parameter to control overall size of image objects. Many studies use a
multiscale approach to deal with the inherently different spectral and spatial
characteristics of disparate features such as buildings, roads, and forests within
a scene. Some studies have tried to identify optimum scales in OBIA. Selecting
optimum scale is often challenging as there is no standardized method to identify
the optimality; many studies simply adopt visual interpretation of objects or the
accuracy of subsequent analyses such as classification as selection criteria. In
addition, scales in most segmentation algorithms are arbitrary values and provide
only a relative comparison between the scales when the same segmentation
algorithm is used to the same input data. There should be standardized guidelines
regarding scales to facilitate the generalization of OBIA in remote sensing
applications, to enable efficient comparison of different OBIA approaches, and
to select optimum scales. The main objectives of this chapter are to provide a
review of recent publications (since 2008) in OBIA focusing on scale optimization
and discuss the related trends.
10.2 BRIEF OVERVIEW
We focused on peer-reviewed papers on OBIA published since 2008. The journals
we searched for papers include Remote Sensing of Environment (19), Photogrammetric Engineering and Remote Sensing (15), International Journal of Applied
Earth Observation and Geoinformation (8), ISPRS Journal of Photogrammetry and
Remote Sensing (8), IEEE Transactions on Geoscience and Remote Sensing (3),
International Journal of Remote Sensing (4), and others (19). We reviewed a total
of 76 papers, and Figures 10.1–10.3 provide a brief overview of the selected papers
for review. The number of publications has generally increased over the past five
years (Figure 10.1). Most of the selected studies utilized the commercial software
tool eCognition for performing the image segmentation that is a key step in OBIA
BRIEF OVERVIEW
199