functions and a hierarchical approach. Whiteside et al. (2011) compared object- and
pixel-based classification for mapping savannas from ASTER data using nearestneighbor and fuzzy classification methods. Multiscale image segmentation was used
to generate both coarse and fine objects.
Data from active sensors such as radar and lidar have often been used in OBIA
applications (Liu J. et al., 2008; Qi et al., 2012). Arnesen et al. (2013) used multiscale
segmentation OBIA to map flood condition by land cover type using the ALOS
ScanSAR backscattering signature. Dră gut and Eisank (2011) classified topography
from Shuttle Radar Topography Mission (SRTM) data using OBIA. Three scale
levels that represent different complexities based on self-adaptive and data-driven
techniques were used for topography classification. Im et al. (2008a) introduced an
OBIA approach based solely on the analysis of lidar-derived information for land
cover classification.
As mentioned in the Section 3.1, object-based change detection is a current focus
in OBIA and different change detection approaches have been employed (Conchedda et al., 2008; de Chant and Kelly, 2009; Dronova et al., 2011; Im et al., 2008b;
Stow et al., 2008; Wang et al., 2009). Some studies integrated other approaches with
OBIA. For example, Bontemps et al. (2008) proposed an automated object-based
change detection method based on a probabilistic changed–unchanged threshold
procedure from time-series images. The differentiation between inter- and intraannual dynamics calculated in the method provides efficiency when identifying
changes associated with natural variability such as the phenological cycle of
vegetation. Doxani et al. (2012) investigated urban changes based on scale-space
filtering embedded in OBIA and multivariate alteration detection (MAD) transformation. Selection of optimum scale for object-based change detection is crucial for
successful change detection. Dronova et al. (2012) used OBIA to classify the Poyang
Lake wetland in China. They determined an optimum segmentation scale using a
trial-and-error method and optimized class discrimination based on machine learning
approaches.
10.3.2 Algorithm Oriented
Scientists have continued to improve OBIA approaches by fusing different algorithms
(Mahmoudi et al., 2013; Sebari and He, 2013). Cai et al. (2009) proposed a
segmentation method based on a watershed transformation and multiscale merging.
Pan et al. (2009) developed an improved strategy for image segmentation based on the
mean-shift segmentation and the fractal net evolution approach (FNEA). Their
method reduced the number of objects compared to traditional FNEA, which would
benefit the subsequent classification. Tzotsos et al. (2011) described an OBIA
approach that integrated nonlinear scale-space filtering into a multiscale segmentation
and classification procedure. Their approach does not require testing parameters such
as shape, color, and texture that control the segmentation process. OBIA has also been
combined with a subpixel mapping approach to map buildings with prior knowledge
of building shape (Ling et al., 2012).
202
OPTIMUM SCALE IN OBJECT-BASED IMAGE ANALYSIS
pixel-based classification for mapping savannas from ASTER data using nearestneighbor and fuzzy classification methods. Multiscale image segmentation was used
to generate both coarse and fine objects.
Data from active sensors such as radar and lidar have often been used in OBIA
applications (Liu J. et al., 2008; Qi et al., 2012). Arnesen et al. (2013) used multiscale
segmentation OBIA to map flood condition by land cover type using the ALOS
ScanSAR backscattering signature. Dră gut and Eisank (2011) classified topography
from Shuttle Radar Topography Mission (SRTM) data using OBIA. Three scale
levels that represent different complexities based on self-adaptive and data-driven
techniques were used for topography classification. Im et al. (2008a) introduced an
OBIA approach based solely on the analysis of lidar-derived information for land
cover classification.
As mentioned in the Section 3.1, object-based change detection is a current focus
in OBIA and different change detection approaches have been employed (Conchedda et al., 2008; de Chant and Kelly, 2009; Dronova et al., 2011; Im et al., 2008b;
Stow et al., 2008; Wang et al., 2009). Some studies integrated other approaches with
OBIA. For example, Bontemps et al. (2008) proposed an automated object-based
change detection method based on a probabilistic changed–unchanged threshold
procedure from time-series images. The differentiation between inter- and intraannual dynamics calculated in the method provides efficiency when identifying
changes associated with natural variability such as the phenological cycle of
vegetation. Doxani et al. (2012) investigated urban changes based on scale-space
filtering embedded in OBIA and multivariate alteration detection (MAD) transformation. Selection of optimum scale for object-based change detection is crucial for
successful change detection. Dronova et al. (2012) used OBIA to classify the Poyang
Lake wetland in China. They determined an optimum segmentation scale using a
trial-and-error method and optimized class discrimination based on machine learning
approaches.
10.3.2 Algorithm Oriented
Scientists have continued to improve OBIA approaches by fusing different algorithms
(Mahmoudi et al., 2013; Sebari and He, 2013). Cai et al. (2009) proposed a
segmentation method based on a watershed transformation and multiscale merging.
Pan et al. (2009) developed an improved strategy for image segmentation based on the
mean-shift segmentation and the fractal net evolution approach (FNEA). Their
method reduced the number of objects compared to traditional FNEA, which would
benefit the subsequent classification. Tzotsos et al. (2011) described an OBIA
approach that integrated nonlinear scale-space filtering into a multiscale segmentation
and classification procedure. Their approach does not require testing parameters such
as shape, color, and texture that control the segmentation process. OBIA has also been
combined with a subpixel mapping approach to map buildings with prior knowledge
of building shape (Ling et al., 2012).
202
OPTIMUM SCALE IN OBJECT-BASED IMAGE ANALYSIS