). Beijing 2008 
L 
BUILDING EXTRACTION BASED ON 
DENSE STEREO MATCH AND EDISON ALGORITHM 
BingXuan GUO \ ZhiChao ZHANG a , YuanZheng SHAO a , Qi Li a 
a State Key Lab of Information Engineering in Surveying Mapping and Remote Sensing,Wuhan University, China 
-mobilemap@gmail.com 
Commission WGS-PS, WG III/4 
KEY WORDS: Extraction, Digital, Photogrammetry, Semi-automation, Dense Stereo Match, EDISON, Building Extraction 
adapt the cur 
ii. Uray et al. 
i be expanded 
d from surface 
which we can 
. (Hong et al., 
for increasing 
ABSTRACT: 
Based on the high-resolution image data in urban areas, the paper investigate the key technologies in building detection and 
reconstruction, particularly focused on the Building Model Driving (BMD) algorithm for hip-roof building and box building. In this 
case, the boundaries of building rooftop are straight lines. The main methods, including generating the DSM data by a graph-cut 
based image dense stereo match under the energy minimization framework, constructing the coarse building model by the BMD 
algorithm, obtaining the 2D Building Edge Feature Vector (BEFV) by the EDISON algorithm. In order to refine the coarse model, 
projecting the coarse 3D building model to the original image and making a constraint to the edge of building rooftop by the BEFV. 
An integration has been made under all the above method to generate a semi-automatic building extracting system. 
eTRIMS, E - 
nes. 
. Eigenfaces vs. 
ejection. PAMI 
itropy approach 
; generative vi- 
lental Bayesian 
t Workshop on 
lative Subspace 
r Subsampling. 
a clustering by 
symmetry and 
n from a single 
tomatic Object 
i: CVPR. 
it visual search 
an, LNCS, Voi. 
!006. Why to 
? or incremental 
iblic. 
rdis, A., 2007. 
ctive and Dis- 
’007. Towards 
deh, Germany, 
1. INTRODUCTION 
Digital image acquisition of high resolution, accuracy and 
multi-spectrum and real-time imagery is coming true, how 
quickly and efficiently to survey the interested targets and 
obtain the information from these images has become an active 
research topic in digital photogrammetry. Till now, although 
many experts of diverse background have struggled for many 
years to create a automatic building extraction system, (Baillard, 
C., Schmid, C, 1999) there is still no one adaptive enough to 
handle most remotely sensed data. So building extraction semi- 
automatically from the high-resolution images is a more 
practical aspect. The pattern recognition ability of human in 
conjunction with the high process speed of computer can make 
a more practical building extraction system in the foreseeable 
future. 
The semi-automatically extraction efficiency and accuracy of 
the building model from the images is limited for the lack of the 
excellent building model reconstruction algorithm. Previous 
works has been done in this field, a excellent review of 
detection of buildings from aerial space images has been made 
by Prof.Grun, etc (Grun et al, 1995; Robert W. Carroll), some 
more work is described in (Fua, 1996; Henricsson et al, 1996; 
Weidner, 1996). In recent years, more and more new kinds of 
inputs, such as range images and spectrum images are used. In 
this paper, we still focus on the use of stereo images. 
In this paper, the reconstruction of buildings is split into three 
steps: in the first step the origin stereopair are preprocessed, 
including DSM generation, BMD algorithm model extraction. 
Then a coarse model is recovered, which consists of the three- 
dimensional edges of the building and the height of the building. 
In the final step, the 3D edges of the coarse model are 
reprojected to the original images to refine the model. In the 
following, the four steps are described in detail. 
In Sections 2 the related technologies and concept , such as 
EDISON algorithm, disparity space image and dense stereo 
match, are considered. Section 3 describes the strategy of the 
Building Model Driving (BMD) algorithm. Section 4 presents 
the systematic integration which illustrates the full process. A 
proving experiment is also performed in section 5 and the 
conclusion is drawn in section 6. 
2. BACKGROUND 
2.1 EDISON algorithm 
As the edge detection is one of the bases of our buildings 
extraction algorithm, we made much more efforts on it. After 
many tests and comparisons, we introduce EDISON algorithm 
to detect edge of the images. (P. Meer, B. Georgescu, 2001) 
With few exceptions, the fundamental assumption of all step 
edge detectors is that the regions on either side of an edge are 
constant in colors or intensity. Much more effort has gone into 
making them robust to noise, but the noise is assumed to have 
statistically simple properties. 
Convolution masks are ideal for realizing such assumption 
because the sign of the weight at a pixel tells us what side of the 
edge it is hypothesized to be on. We can think of a convolution 
as finding the weighted mean of each side and then computing 
the distance between the two means (Eric N. Mortensen, 2001; 
Mark A. Ruzon, 1999).While this assumption holds well 
enough for many applications, it does not hold in all cases. For 
instance, as scale increases, it is more likely that the weighted 
mean of each side will not be meaningful because an operator 
will include image features unrelated to the edge. This 
observation is even truer of color images. When only intensities 
are involved, the average over a large window is still 
perceptually meaningful because intensities are totally ordered. 
In color images, there is no such ordering, so the “mean color” 
of a large window may have little perceptual similarity to any 
of the colors in it. 
15. Symmetry- 
In: CVIU, 
405