relationships between a chlorophyll spectral index and vegetation chlorophyll contents at the leaf, canopy, and landscape scales. Significant relationships are found at all
three scales, suggesting that it is feasible to accurately estimate chlorophyll contents
using both ground and space remote sensing data.
In remote sensing, image segmentation has a longer history and has its roots in
industrial image processing but was not used extensively in the geospatial community in
the 1980s and 1990s (Blaschke, 2010). Object-oriented image analysis has been
increasingly used in remote sensing applications due to the advent of high-spatialresolution image data and the emergence of commercial software such as eCognition
(Benz et al., 2004; Wang et al., 2004). In the process of creating objects, a scale
determines the occurrence or absence of an object class. Thus, the issue of scale and
scaling are fundamental considerations in the extraction, representation, modeling, and
analyses of image objects (Hay et al., 2002; Tzotsos et al., 2011).
The three chapters in Part III focus on discussion of these issues. In Chapter 8, Hay
introduces a novel geo-object-based framework that integrates hierarchy theory and
linear scale space (SS) for automatically visualizing and modeling landscape scale
domains over multiple scales. Specifically, this chapter describes a three-tier hierarchical methodology for automatically delineating the dominant structural components
within 200 different multiscale representations of a complex agro-forested landscape.
By considering scale-space events as critical domain thresholds, Hay further defines a
new scale-domain topology that may improve querying and analysis of this complex
multiscale scene. Finally, Hay shows how to spatially model and visualize the
hierarchical structure of dominant geo-objects within a scene as “scale-domain manifolds” and suggests that they may be considered as a multiscale extension to the
hierarchical scaling ladder as defined in the hierarchical patch dynamics paradigm.
Chapter 9 by Tzotsos, Karantzalos, and Argialas introduces a multiscale object-oriented
image analysis framework which incorporates a region-merging segmentation algorithm enhanced by advanced edge features and nonlinear scale-space filtering. Initially,
edge and line features are extracted from remote sensing imagery at several scales using
scale-space representations. These features are then used by the enhanced segmentation
algorithm as constraints in the growth of image objects at various scales. Through
iterative pairwise object merging, the final segmentation can be achieved. Image objects
are then computed at various scales and passed on to a kernel-based learning machine
for classification. This image classification framework was tested on very high
resolution imagery acquired by various airborne and spaceborne panchromatic, multispectral, hyperspectral, and microwave sensors, and promising experimental results
were achieved. Chapter 10, by Im, Quackenbush, Li, and Fang, provides a review of
recent publications on object-based image analysis (OBIA) focusing on determination
of optimum scales for image segmentation and the related trends. Selecting optimum
scale is often challenging, since (1) there is no standardized method to identify the
optimality and (2) scales in most segmentation algorithms are arbitrarily selected. The
authors suggest that there should be transferable guidelines regarding segmentation
scales to facilitate the generalization of OBIA in remote sensing applications, to enable
efficient comparison of different OBIA approaches, and to select optimum scales for the
multitude of different image components.
CHARACTERIZING, MEASURING, ANALYZING, AND MODELING SCALE
7
three scales, suggesting that it is feasible to accurately estimate chlorophyll contents
using both ground and space remote sensing data.
In remote sensing, image segmentation has a longer history and has its roots in
industrial image processing but was not used extensively in the geospatial community in
the 1980s and 1990s (Blaschke, 2010). Object-oriented image analysis has been
increasingly used in remote sensing applications due to the advent of high-spatialresolution image data and the emergence of commercial software such as eCognition
(Benz et al., 2004; Wang et al., 2004). In the process of creating objects, a scale
determines the occurrence or absence of an object class. Thus, the issue of scale and
scaling are fundamental considerations in the extraction, representation, modeling, and
analyses of image objects (Hay et al., 2002; Tzotsos et al., 2011).
The three chapters in Part III focus on discussion of these issues. In Chapter 8, Hay
introduces a novel geo-object-based framework that integrates hierarchy theory and
linear scale space (SS) for automatically visualizing and modeling landscape scale
domains over multiple scales. Specifically, this chapter describes a three-tier hierarchical methodology for automatically delineating the dominant structural components
within 200 different multiscale representations of a complex agro-forested landscape.
By considering scale-space events as critical domain thresholds, Hay further defines a
new scale-domain topology that may improve querying and analysis of this complex
multiscale scene. Finally, Hay shows how to spatially model and visualize the
hierarchical structure of dominant geo-objects within a scene as “scale-domain manifolds” and suggests that they may be considered as a multiscale extension to the
hierarchical scaling ladder as defined in the hierarchical patch dynamics paradigm.
Chapter 9 by Tzotsos, Karantzalos, and Argialas introduces a multiscale object-oriented
image analysis framework which incorporates a region-merging segmentation algorithm enhanced by advanced edge features and nonlinear scale-space filtering. Initially,
edge and line features are extracted from remote sensing imagery at several scales using
scale-space representations. These features are then used by the enhanced segmentation
algorithm as constraints in the growth of image objects at various scales. Through
iterative pairwise object merging, the final segmentation can be achieved. Image objects
are then computed at various scales and passed on to a kernel-based learning machine
for classification. This image classification framework was tested on very high
resolution imagery acquired by various airborne and spaceborne panchromatic, multispectral, hyperspectral, and microwave sensors, and promising experimental results
were achieved. Chapter 10, by Im, Quackenbush, Li, and Fang, provides a review of
recent publications on object-based image analysis (OBIA) focusing on determination
of optimum scales for image segmentation and the related trends. Selecting optimum
scale is often challenging, since (1) there is no standardized method to identify the
optimality and (2) scales in most segmentation algorithms are arbitrarily selected. The
authors suggest that there should be transferable guidelines regarding segmentation
scales to facilitate the generalization of OBIA in remote sensing applications, to enable
efficient comparison of different OBIA approaches, and to select optimum scales for the
multitude of different image components.
CHARACTERIZING, MEASURING, ANALYZING, AND MODELING SCALE
7