Refining High Spatial Resolution Remote
Sensing Image Segmentation for Man-made
Objects through aCollinear and Ipsilateral
Neighborhood Model
Min Wang, Yanxia Sun, and Guanyi Chen
Abstract
Man-made objects, such as buildings and roads, which are
important targets for information extraction from high spatial
resolution (
HSR
) remote sensing images, often feature straight
boundaries. This study employs this knowledge on HSR image
segmentation by embedding a straight-line constraint in region-
based image segmentation. A new concept called collinear and
ipsilateral neighborhood is proposed and applied to hard-
boundary constraint-based image segmentation for accuracy
improvement. In the experimental areas, the method accuracy
measured by recall ratio r increases from 0.036 to 0.048 (on
the average) after the refinement, with significantly smaller
decreases in precision p that are all less than 0.006. In sum, the
proposed technique effectively reduces over-segmentation errors
and maintains the same level of under-segmentation error ratio,
particularly in man-made areas. It facilitates subsequent object-
based image analyses, including feature extraction, object
recognition, and classification.
Introduction
Image segmentation is a most important step in object-based
image analysis (
OBIA
); it significantly influences the succeeding
steps, including feature extraction and classification. The pio-
neer
OBIA
software is eCognition
(originally developed at Defini-
ens AG) by Trimble Inc. (2014), which features a multi-resolu-
tion segmentation method (fractal net evolution approach or
FNEA
). This method has wide applicability, high efficiency, and
high accuracy. However,
FNEA
needs to be improved in terms of
under- and over-segmentation error ratios, input dependency,
and segment boundary precision (Wang and Li, 2014).
In our previous study (Wang and Li, 2014), we proposed a
novel segmentation method based on a hard-boundary con-
straint and two-stage merging (
HBC-SEG
). This novel method
exhibits improved performance when compared with
FNEA
.
In the current study, we design a refined
HBC-SEG
by integrat-
ing straight-line constraints because man-made objects in
HSR
images often have straight boundaries. The main contributions
of this work are as follows. First, we propose and implement
a range of techniques, including extracting two types of object
primitives (OPs) (segments and straight lines), building their
mutual spatial topologies, and comprehensively utilizing these
OPs in image segmentation. Second, we propose a new neigh-
borhood model called collinear and ipsilateral neighborhood
(Without ambiguity, we refer to it as
IPSL
-neighborhood). We
then confirm that
IPSL
-neighborhood can improve segmentation
accuracy in
HSR
images. The proposed model and technique
are significant to
OBIA
considering that man-made objects are
often the main target of
HSR
image information extraction.
The rest of this paper is organized as follows. The next
section provides a review of related work on remote sensing
image segmentation, followed by a detailed discussion of the
proposed method, including a brief introduction of
HBC-SEG
,
straight-line primitive extraction, line and segment topology
modeling, and refined segmentation method. The next section
presents the experiments conducted, followed by a summary
of this study.
Related Work
Image segmentation aims to partition an image into several
segments, such that each segment is homogeneous, but none
of the unions of two adjacent segments is homogeneous (Pal
and Pal, 1993). Segmentation accuracy can be measured
based on over- and under-segmentation. Over-segmentation
indicates that a homogeneous region is divided into several
segments, whereas under-segmentation means that different
regions are grouped into one segment. Current remote sensing
image segmentation methods include point/pixel-based, edge-
based, region-based, texture-based, and hybrid. Previous stud-
ies (Pal and Pal, 1993; Schiewe, 2002; Shankar, 2007; Dey
et
al
., 2010) have provided systematic reviews of these methods.
In the field of remote sensing applications,
FNEA
(Baatz
and Shäpe, 2000), along with the successful business appli-
cation of eCognition
software, is the most popular segmen-
tation method for
OBIA
. As an important algorithm param-
eter, scale is utilized to control the average segment size in
segmentations. From scale changes (small to large), segments
are merged gradually and hierarchically to allow for multi-
resolution segmentation. However, the global scale parameter
is limited because remote sensing images contain different
types of large and small ground objects. Most ground objects
may be over-segmented at small scales. Several small objects
may be under-segmented if large scales are specified, whereas
several large objects may remain over-segmented. Thus, the
co-existence of over- and under-segmentation often results in
manual scale tuning to minimize and balance segmentation
errors, which are tedious.
Key Laboratory of Virtual Geographic Environment (Nanjing
Normal University), Ministry of Education, Nanjing, Jiangsu,
P.R. China, 210023; and the Jiangsu Center for Collaborative
Innovation in Geographical Information Resource
Development and Application, Nanjing, Jiangsu, P.R. China,
210023 (
).
Photogrammetric Engineering & Remote Sensing
Vol. 81, No. 5, May 2015, pp. 397–406.
0099-1112/15/397–406
© 2015 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.81.5.397
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
May 2015
397
