IEEE Geoscience and Remote Sensing Magazine - December 2015 - 23

distinctively internal spatial distribution of
heterogeneous units. The roof structure is
described not only by a conventional spectrum histogram and regional area index but
also by the Euler number matrix representing heterogeneous spatial distribution. Corresponding similarity measures are then
introduced. The overall methodology is
presented schematically in Fig. 1.
2.1 Regional featuRe detection
Feature-based matching relies on the robustness and distinctiveness of detected
features. We segment the features of roofs
and classify heterogeneous regions at the
same time, thereby constructing a regional
feature that represents the building. Thus,
the feature contains abundant information,
rectangular contour, internal structure, and
original spectrum.

Original Images
External Contour

Internal Structure

EDLines

K-Means

Line Segments

Classified Map

Perceptual Organization

Binarization

Rectangulars

Binary Image

Optimization

Morphology
Median Filtering

Feature Contour

Internal Structure

Regional Feature
Area Index

Topological Matrix
Similarity
A. ExtrActing ExtErior outlinEs
Gray Histogram
The buildings cannot be simply identified
by their spectral and textural features in
Matched Images
very high resolution optical remote sensing images because of the differences in
materials, size, shape and sensors. How- Figure 1. The framework of the proposed procedure consisted of regional feature
ever, urban buildings are characteristic of extraction and similarity measuring.
regular geometry (especially the rectangle
edges are split into several fractions because of the imshape), and thus buildings with rectangular structure are
age noise and complex surface scene, for which recomwidely and well extracted based on line segments detecbination is necessary. Least square fitting is introduced
tion and perceptual grouping. Generally, line segments
to replace conventional linear organization and restore
are detected and then recombined to straight lines firstly,
the edges, thus reducing computational complexity. 3)
and perceptual grouping extracts right-angle structures
The right-angle structure is then detected on the basis of
based on the spatial relationship of straight lines and
the distance and angle relationship. In our process, the
consequently rectangular structures upon the spatial
T-shaped structure caused by a common boundary and
relationship of right-angle structures. Liu[15] made an
the double-bounded structure caused by a shadow are
outstanding contribution to ascertain undetected edges
fully considered by utilizing a developed splitting-up
(Fig. 2 (left)) of buildings with fuzzy edges by establishmethod and buffer analysis, respectively. 4) The opening an elastic rectangle template (Fig. 2 (right)), which
ing direction of the right-angle structure is then merged
would automatically complete rectangle structure acwith the endpoints to construct a U-shaped structure.
cording to image gradient (Fig. 3).
5) Rectangles are consequently constructed by combinIn the paper, we extract the rectangular building roof
ing the opposite U-shaped structure if one exists. Othutilizing a newly line segment detector, a stable linear
erwise, an elastomeric rectangular template is used to
organization and an elastic template, which mainly consists of six steps: 1) An advanced straight line segment
detection algorithm called EDLines [16] is implemented.
This algorithm is suitable for automatic high-precision
matching because the linear time line segment detector requires no parameter tuning, provides continuous,
clean, and accurate results, and controls false linear features soundly. EDLines is robust against scale and view
variation compared with the typical Hough transformation and LSD algorithm [17]. 2) The discrete line segFigure 2. Building in images with several vague edges and overments are reconnected based on their spatial relationlapping line detection results (Left); Rectangular elastic template
ships, including distance and angle constraints. The
schematic diagram (Right).
december 2015

ieee Geoscience and remote sensing magazine

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - December 2015