Multi-resolution and multi-spectral image fusion for
urban object extraction
Yun Zhang and Ruisheng Wang
Department of Geodesy and Geomatics Engineering
University of New Brunswick
15 Dineen Dr., Fredericton, New Brunswick, CANADA E3B 5A3
YunZhang@UNB.ca; «Ruisheng Wang? i2z51(gunb.ca
XX"' ISPRS Congress
KEYWORDS: Digital, Urban, Object, Multispectral, Fusion, Classification, Extraction
ABSTRACT:
A new approach for object extraction from high-resolution satellite images is presented in this paper. The new approach integrates
image fusion, multi-spectral classification, feature extraction and feature segmentation into the object extraction of high-resolution
satellite images. Both spectral information from multispectral (MS) images and spatial information from panchromatic (Pan) images
are utilized for the extraction to improve accuracies. This paper mainly concentrates on road extraction from QuickBird MS and Pan
images using the proposed approach. Experiments of road extraction with QuichBird MS and Pan images demonstrate that the
proposed approach is effective. The completeness and correctness of road network extraction reaches 0.95, significantly higher than
those of other existing road extraction methods.
1. INTRODUCTION
The latest high-resolution satellites, such as IKONOS,
QuickBird and OrbView, all simultaneously collect multi-
spectral (MS) images at a lower resolution (4m or 2.8m) and
panchromatic (Pan) images at a higher resolution (1m or 0.7m).
In the multi-spectral images, most urban objects can be visually
recognized according to the colour difference, but they cannot be
clearly delineated due to the lack of spatial resolution. In the
panchromatic images, however, the shape of most individual
objects can be clearly identified, but many of them cannot be
classified due to the lack of spectral information.
To extract useful information from available high-resolution
images, including airborne and spaceborne imagery, different
automatic and semi-automatic approaches have been developed.
To date, automatic techniques for information extraction from
imagery can be divided into two main categories: (1) multi-
spectral classification techniques to classify objects from multi-
spectral images, and (2) grey value and feature based techniques
to extract objects from panchromatic images.
Multi-spectral classification techniques are efficient for
classifying homogeneous objects. But they are inefficient for the
classification of urban objects when high-resolution images are
used, because many urban objects are not homogeneous in
colour and different objects may appear in similar or same
colour, such as asphalt paved roofs and roads. Further, the
classification results usually are coarse due to the lower
resolution of the multispectral images.
On the other hand, grey value and feature based techniques have
been successful in extraction of objects from some simple
panchromatic scenes. But they are still very limited in the
extraction of urban objects due to the complexity of urban
scenes and due to the lack of spectral information. In urban areas
different objects may appear in similar grey values and have
960
similar shapes, such as building roofs and small parking lots,
so that they can be hardly differentiated automatically
according to the grey values and shapes.
In this paper, we present a new approach to better extract
urban objects, which utilizes both spectral information from
multi-spectral images and spatial information from
panchromatic images. Different from existing techniques, this
new approach effectively integrates techniques of image
fusion, multi-spectral classification, and feature extraction into
the extraction process. New image fusion and edge-aided
classification algorithms and software have been developed.
The concept of the new approach can be used for urban road
network extraction, building extraction, and other object
extractions. Here we mainly concentrate on road extraction
using QuickBird multispectral and panchromatic images. The
road extraction results demonstrated that this new approach is
significantly superiors to the conventional multi-spectral
classification, panchromatic image based feature extraction,
and other existing multi-spectral and panchromatic image
integrated classifications.
2. THE PROPOSED APPROACH
Figure 1 illustrates the general process of the proposed
approach. To overcome the shortcoming of classification of
low-resolution MS images, the MS and Pan QuickBird images
are first fused into a pan-sharpened MS image. An
unsupervised classification is then applied to the pan-
sharpened image to obtain a classified road image. And an
edge detection approach is applied to the Pan image to obtain
an edge image. In the edge-aided segmentation, the binary
edge image from the Pan image is employed to segment the
classified road image from the pan-sharpened image. Then, a
shape-based segmentation and a segments filtering algorithm
are employed to remove non-road objects.
Inte:
—
Fig
dez
pro
fol