Many studies have investigated optimum scale selection in algorithm-oriented
OBIA studies (Aguirre-Gutiérrez et al., 2012; Laliberte and Rango, 2009). Tong
and Li (2011) investigated the influence of shape parameters on selecting an
optimal scale value in multiresolution segmentation using QuickBird imagery.
Tong and Li (2011) tested 26 parameter settings for segmentation, with each setting
having 10 segmentation scales. They found that different shape parameters had
different impacts on the optimum scale and a greater weight on the shape parameter
might improve the efficiency of segmentation. Yi et al. (2012) proposed a flexible
scale synthesis method to meet different segmentation requirements in image
analysis. Their method divides the image into multiple regions where each region
contains ground objects at similar scale. They identified the optimal segmentation
scale using the method by Crevier (2008). The final segmentation result is achieved
through synthesis of the optimum segmentation of each region, which yields more
coherent results to ground objects.
Peña-Barragán et al. (2011) investigated 1312 segmentation scenarios with
different parameter settings for object-based crop identification and mapping from
multiseasonal ASTER data. They applied an empirical discrepancy method (Ortiz and
Oliver, 2006; Zhang, 1996) to assess the quality of the segmentation results to identify
the optimum parameter settings including the scale. An adaptive spectral matching
method to extract thematic objects from Landsat ETM+ imagery was employed in
Qiao et al. (2012). Two scales—whole and local—were used for spectral matching
through end-member selection. Liu et al. (2010) used a mean-shift algorithm to
segment digital aerial imagery. Multiple scales were tested and optimum scale
segmentation was selected based on a trial-and-error method. Martha et al. (2011)
optimized segmentation based on a plateau objective function derived from spatial
autocorrelation and intrasegment variance analysis. Their optimum segmentation
results were used in a knowledge-based classification approach to landslide detection.
Anders et al. (2011) proposed a stratified OBIA to map alpine geomorphology from
airborne lidar data. They used two dimensional frequency distribution matrices of
training samples and image objects to identify an optimum scale. Shruthi et al. (2011)
extracted gully erosion features using OBIA from IKONOS and GeoEye-1 data.
Multiple scales were tested and the estimation of scale parameter (ESP) was used to
determine the optimum scale. Drǎ gut et al. (2010) developed ESP to assist in selecting
a suitable range of scales for image segmentation using local variance of object
heterogeneity within a scene.
Some studies focused on developing new segmentation algorithms. Wang et al.
(2010) proposed a new segmentation algorithm, called the region-based image
segmentation algorithm (RISA), based on k-means clustering and region growing
and merging. They evaluated RISA through the quantitative assessment of object
quality and classification accuracy and found that the RISA performance was similar
to the multiscale resolution segmentation algorithm in eCognition. Zhang et al. (2013)
proposed a boundary-constrained multiscale segmentation approach and tested it with
a set of high-resolution images, including QuickBird, WorldView, and aerial image
data. Their approach can produce nested multiscale segmentations and is especially
good at delineating boundaries smoothly.
RECENT STUDIES IN OBIA
203
OBIA studies (Aguirre-Gutiérrez et al., 2012; Laliberte and Rango, 2009). Tong
and Li (2011) investigated the influence of shape parameters on selecting an
optimal scale value in multiresolution segmentation using QuickBird imagery.
Tong and Li (2011) tested 26 parameter settings for segmentation, with each setting
having 10 segmentation scales. They found that different shape parameters had
different impacts on the optimum scale and a greater weight on the shape parameter
might improve the efficiency of segmentation. Yi et al. (2012) proposed a flexible
scale synthesis method to meet different segmentation requirements in image
analysis. Their method divides the image into multiple regions where each region
contains ground objects at similar scale. They identified the optimal segmentation
scale using the method by Crevier (2008). The final segmentation result is achieved
through synthesis of the optimum segmentation of each region, which yields more
coherent results to ground objects.
Peña-Barragán et al. (2011) investigated 1312 segmentation scenarios with
different parameter settings for object-based crop identification and mapping from
multiseasonal ASTER data. They applied an empirical discrepancy method (Ortiz and
Oliver, 2006; Zhang, 1996) to assess the quality of the segmentation results to identify
the optimum parameter settings including the scale. An adaptive spectral matching
method to extract thematic objects from Landsat ETM+ imagery was employed in
Qiao et al. (2012). Two scales—whole and local—were used for spectral matching
through end-member selection. Liu et al. (2010) used a mean-shift algorithm to
segment digital aerial imagery. Multiple scales were tested and optimum scale
segmentation was selected based on a trial-and-error method. Martha et al. (2011)
optimized segmentation based on a plateau objective function derived from spatial
autocorrelation and intrasegment variance analysis. Their optimum segmentation
results were used in a knowledge-based classification approach to landslide detection.
Anders et al. (2011) proposed a stratified OBIA to map alpine geomorphology from
airborne lidar data. They used two dimensional frequency distribution matrices of
training samples and image objects to identify an optimum scale. Shruthi et al. (2011)
extracted gully erosion features using OBIA from IKONOS and GeoEye-1 data.
Multiple scales were tested and the estimation of scale parameter (ESP) was used to
determine the optimum scale. Drǎ gut et al. (2010) developed ESP to assist in selecting
a suitable range of scales for image segmentation using local variance of object
heterogeneity within a scene.
Some studies focused on developing new segmentation algorithms. Wang et al.
(2010) proposed a new segmentation algorithm, called the region-based image
segmentation algorithm (RISA), based on k-means clustering and region growing
and merging. They evaluated RISA through the quantitative assessment of object
quality and classification accuracy and found that the RISA performance was similar
to the multiscale resolution segmentation algorithm in eCognition. Zhang et al. (2013)
proposed a boundary-constrained multiscale segmentation approach and tested it with
a set of high-resolution images, including QuickBird, WorldView, and aerial image
data. Their approach can produce nested multiscale segmentations and is especially
good at delineating boundaries smoothly.
RECENT STUDIES IN OBIA
203