(Figure 10.2). As expected, OBIA has been more commonly applied to highspatial-resolution data (Figure 10.3). In Section 10.3, we review these papers
focusing on what scales were used in image segmentation and how optimum scale
was determined.
10.3 RECENT STUDIES IN OBIA
10.3.1 Application Oriented
Land cover classification is perhaps the application field that most frequently applies
OBIA. Many studies investigated land cover and land use information in urban/
suburban areas using OBIA approaches (Durieux et al., 2008; Im et al., 2008a;
Lizarazo and Barros, 2010; Meng et al., 2012; Stow et al., 2010; Tiede et al., 2010;
Zhou et al., 2009). The heterogeneous nature of urban areas typically requires careful
selection of a scale parameter, since this directly influences the size of image objects.
Multiscale approaches are commonly adopted to extract urban features that vary in
terms of color and shape. For example, Jacquin et al. (2008) introduced a hybrid
OBIA approach to map urban sprawl in a periurban environment using SPOT images.
Different scale values were used to extract urban objects at regional and local scales.
They suggested a hybrid OBIA approach that combines multiscales to deal with the
spatial and spectral heterogeneity of the urban surface. Huang and Qi (2010) also
applied a multiscale segmentation to IKONOS data to extract land cover information
using different scales for different land cover types.
As generally agreed, the higher the spatial resolution of an image, the better the
expected segmentation (Holt et al., 2009; Karantzalos and Argialas, 2009; Wilschut
et al., 2013). However, higher spatial resolution does not necessarily yield higher
classification accuracy. For example, Lian and Chen (2011) used OBIA to extract
urban features such as buildings, roads, water bodies, and vegetation from different
resolution images (i.e., two SPOT data at 10 and 2.5 m, ASTER at 15 m, and
QuickBird at 0.6 m) at different scales. They found the QuickBird experiment resulted
in lower classification accuracy than the other experiments. OBIA has been frequently
used with moderate and coarse resolution data such as the Landsat TM and MODIS
(Ranson et al., 2011; Yang et al., 2013; Zhu and Woodcock, 2012).
OBIA studies often adopt a comparison between pixel- and object-based
approaches (Chirici et al., 2011; Johansen et al., 2010; Myint et al., 2011; Verbeeck
et al., 2012; Watmough et al., 2011). For example, Duro et al. (2012) compared pixelbased methods with OBIA to classify agricultural landscapes using selected machine
learning algorithms such as decision trees, random forest, and support vector
machines. Multiple scales were tested and the optimum scale was identified using
an iterative trial-and-error approach. They found OBIA outperformed the pixel-based
methods when the same machine learning approach was used. Quyang et al. (2011)
mapped vegetation in a saltmarsh ecosystem using QuickBird imagery based on both
pixel- and object-based approaches. They reported that the object-based approach
with a two-scale segmentation was superior to the pixel-based one due to membership
RECENT STUDIES IN OBIA
201
focusing on what scales were used in image segmentation and how optimum scale
was determined.
10.3 RECENT STUDIES IN OBIA
10.3.1 Application Oriented
Land cover classification is perhaps the application field that most frequently applies
OBIA. Many studies investigated land cover and land use information in urban/
suburban areas using OBIA approaches (Durieux et al., 2008; Im et al., 2008a;
Lizarazo and Barros, 2010; Meng et al., 2012; Stow et al., 2010; Tiede et al., 2010;
Zhou et al., 2009). The heterogeneous nature of urban areas typically requires careful
selection of a scale parameter, since this directly influences the size of image objects.
Multiscale approaches are commonly adopted to extract urban features that vary in
terms of color and shape. For example, Jacquin et al. (2008) introduced a hybrid
OBIA approach to map urban sprawl in a periurban environment using SPOT images.
Different scale values were used to extract urban objects at regional and local scales.
They suggested a hybrid OBIA approach that combines multiscales to deal with the
spatial and spectral heterogeneity of the urban surface. Huang and Qi (2010) also
applied a multiscale segmentation to IKONOS data to extract land cover information
using different scales for different land cover types.
As generally agreed, the higher the spatial resolution of an image, the better the
expected segmentation (Holt et al., 2009; Karantzalos and Argialas, 2009; Wilschut
et al., 2013). However, higher spatial resolution does not necessarily yield higher
classification accuracy. For example, Lian and Chen (2011) used OBIA to extract
urban features such as buildings, roads, water bodies, and vegetation from different
resolution images (i.e., two SPOT data at 10 and 2.5 m, ASTER at 15 m, and
QuickBird at 0.6 m) at different scales. They found the QuickBird experiment resulted
in lower classification accuracy than the other experiments. OBIA has been frequently
used with moderate and coarse resolution data such as the Landsat TM and MODIS
(Ranson et al., 2011; Yang et al., 2013; Zhu and Woodcock, 2012).
OBIA studies often adopt a comparison between pixel- and object-based
approaches (Chirici et al., 2011; Johansen et al., 2010; Myint et al., 2011; Verbeeck
et al., 2012; Watmough et al., 2011). For example, Duro et al. (2012) compared pixelbased methods with OBIA to classify agricultural landscapes using selected machine
learning algorithms such as decision trees, random forest, and support vector
machines. Multiple scales were tested and the optimum scale was identified using
an iterative trial-and-error approach. They found OBIA outperformed the pixel-based
methods when the same machine learning approach was used. Quyang et al. (2011)
mapped vegetation in a saltmarsh ecosystem using QuickBird imagery based on both
pixel- and object-based approaches. They reported that the object-based approach
with a two-scale segmentation was superior to the pixel-based one due to membership
RECENT STUDIES IN OBIA
201