PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
November 2013
1
Abstract
Automatic building extraction is currently an important
research topic in the field of photogrammetry. An active
contour model is a well-received approach in this field.
This paper proposes an improved active contour model that
focuses on building extraction from aerial images and lidar
data. The main research concern in this paper is the develop-
ment of energy functions to the optimum use of expert human
knowledge in the overall process. Based on this approach, a
new fuzzy inference system for evaluating energy functions
was developed by modeling the human perception of various
effective parameters in the energy functions. Compared to the
existing active contour models, the new algorithm is capable
of directing the initial contour to building feature boundaries
more quickly and robustly. Accuracy assessment showed that
the proposed model is capable of achieving a shape accuracy
of 98 percent and a total accuracy of 97 percent in complex
urban areas.
Introduction
Urban feature extraction is currently one of the most dif-
ficult and complicated processes that machine vision and
photogrammetry experts are addressing. Information con-
cerning urban features is necessary for various applications
such as urban planning and development,
GIS
(Geographic
Information System) database updates, and urban model
generation. In conventional methods, feature extraction from
digital images is conducted manually, slow, and expensive
process requiring professional operators. The development
of automatic approaches have been considered for the past
decade.
Buildings are considered a vital group of urban features
by researchers in this area. The extraction of building features
from digital images is more complicated, because buildings
have different shapes, roof composite materials, and radio-
metric characteristics.
Numerous methods and algorithms for the
2
D
and
3
D
reconstruction of building models using various sources of
information are reported in the literature. In Dash (2004), the
height variation of objects was used to develop an approach
to separate buildings and vegetation based on the standard
deviation of elevation. Hongjian (2005) extracted and joined
the edge pixels and then computed the building elevation
by using laser scanner data to reconstruct the
3
D
building
model. The fusion of Ikonos satellite imagery and aerial laser
scanner data was used by Sohn (2007) to extract buildings
automatically. An automatic method for building extraction
from the digital elevation model was also proposed by Lafarge
et al.
(2008). Tanathong
et al.
(2009) presented a robust
K.N.Toosi University of Technology, Tehran,
Iran (
).
Photogrammetric Engineering & Remote Sensing
Vol. 79, No, 11, November 2013, pp. 000–000.
0099-1112/13/7911–0000/$3.00/0
© 2013 American Society for Photogrammetry
and Remote Sensing
rectangular building detection process that can discover
small-size buildings in residential areas and large-size build-
ings in industrial areas. Huang and Zhang (2011) proposed
a new morphological building index for automatic building
extraction. By using this index, the relationship between
the implicit characteristics of the building and the proper-
ties of morphological operators has been determined. Huang
et al.
(2011) also investigated information fusion approaches
with high-resolution aerial images and elevation data from
lidar (Light Detection and Ranging) for urban environment
mapping. You and Lin (2011) proposed a novel approach for
building extraction from lidar data. In their paper, a tensor
field has been used to represent the geometric features of
lidar points. Meng
et al.
(2012) presented an object-oriented
method to detect residential buildings for a land use map. In
their method, a multidirectional ground filter was first applied
to generate a bare ground surface from lidar data. Then, using
a morphology-based algorithm, the buildings were detected.
Huang and Zhang (2012) proposed a method for building
extraction from high-resolution imagery that aims to allevi-
ate both commission and omission errors for the original
MBI
(morphological shadow index) algorithm. In this paper, the
improvement includes three aspects: (a) a
MSI
(morphological
shadow index) is proposed to detect shadows that are used as
a spatial constraint of buildings, (b) a dual-threshold filtering
is proposed to integrate the information of
MBI
and
MSI
, and
(c) the proposed framework is implemented in an object-based
environment where a geometrical index and a vegetation
index are then used to remove noise from narrow roads and
bright vegetation.
The active contour model proposed by Kass (1988) was a
successful method that was widely used in image processing
tasks such as image segmentation, image monitoring, and
3
D
object reconstruction. An active contour model is an energy
minimization method that directs the contour to features
such as image edges and lines. To illustrate the capabilities
of this model in object extraction, the model was also used
in building extraction by the researchers in this area. Ruther
(2002) proposed a method for semi-automatic building extrac-
tion from aerial images in dense irregular areas. This method
used an appropriate strategy to extract the initial boundaries
of buildings. An improved active contour model was pro-
posed by Peng (2005) for building extraction from single-band
high-resolution imagery. This model represented an improve-
ment over existing models in the initial seed point selection
process as well as the external energy function. In Mayungaa
(2005), a semi-automatic algorithm was presented for building
Automatic Building Extraction Using a
Fuzzy Active Contour Model
Mostafa Kabolizade, Hamid Ebadi, Mehdi Mokhtarzade
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 11, November 2014, pp. 1061–1068.
0099-1112/14/8011–1061
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.11.1061
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
November 2014
1061