finimum
elepiped
on when
different
e image
inherited
different
luencing
2]e-pixel
f pixels
different
pectrally
training
spatially
mber of
) extract
id auto-
to more
training
ion. For
gle-pixel
spatially
has the
| spatial
ime.
process
sses and
n results
/hen the
hms, the
'ampbell
mmonly
'enerally
cation if
gorithms
ilaximum
w good
Figure 2.Overlay classifications
2.3 Texture Analysis to Improve Per-Pixel Image
Classification
Nevertheless, classification errors caused by similar reflection
(wave length) dramatically affect the separation between built-
form and non-built-form areas. It is rare that the classification
accuracy greater than 80% could be achieved by using per-pixel
classification algorithms (Atkinson and Tate, 1999). In this
context, texture analysis can be a good indicator of the presence
of buildings and other objects and they are usually easier to
detect than the often-complex multi-textured objects which
cause them.
In general, texture analysis approaches are used for recognition
and distinction of different spatial characteristics of spatial
arrangement and frequency of tonal variation related to patterns
or phenomena contained in the digital image or the sensor
image. In this sense, texture image analysis is one of useful
approaches for urban class extraction and separation (Wang and
Hanson 2001, Herold, 2003). Zhang (1999), for example,
combined multi-spectral classification and texture filtering for
building detection in the urban area, and suggested that this
approach increases classification accuracy.
Previous works related to texture image have been carried out
into three categories: development and improvement of texture
extraction algorithms, comparison between texture extraction
schemes, and domain application of extracted texture images.
These types of researches are similar to other cases in digital
image processing, such us image classification. The main
methodologies applied are those related to structural, statistical,
stochastic and space-frequency models (Tuceryan & Jain 1998).
Statistical methods analyse the spatial distribution of grey
values by computing local features at each point in the image
and deriving a set of statistics from the distributions of the local
features (Ojala & Pietikdinen 2004). Statistical methods can be
classified into first-order (one pixel), second-order (two pixels)
and higher-order (three or more pixels). Most frequently used
texture analysis is Grey Level Co-occurrence Matrix (GLCM)
based on second order statistic.
Our research group has developed extensive experience in the
use of per pixel and texture analysis in classification of high
resolution imagery (Alhaddad, Burns & Roca, 2007). The study
of historical images for monitoring urban sprawl (Alhaddad,
Roca & Burns, 2009) has led us to use GeoGraphic Imager
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
plugin, which allows to georeference old images, and, thus,
exploit the capabilities of Photoshop software for image
processing (Morgan & Coops, 2010; Yang, 2009; Gamache,
2007; Gleason, 2007; Peterson & Kelso, 2004).
This methodology, using low resolution images (as Landsat 7)
to detect artificialized land of megacities can be summarized as
follows:
ny ©
General diagram of the gy applied to the classificati gacities {| Sensing].
image processing image Correcting "Artificial"
ENVLEX {or ENVIZOOMI and
EM Phatoshon Gesgrazhic imager
Pixel Based Object Based
ispectrai - wavelength! (texture, form...)
Ts i i: xi
d
D i Artificial
Final Result
RU DS
iMac Cocabat futiffigal 11 n
suus
Based
EHVLEX tes
Be Nun :
Lusi Bsn
| FNAL RESULT
Figure 3.General Diagram of the Methodology
3. ANALYSIS OF THE RESULTS
3.1 Built-up Land of Mega-Cities
The above methodology, applied to the eight mega-cities
(Tokyo, Mexico, Chicago, Moscow, New York, London, Sao
Paulo and Shanghai), allows the delineation of the built up land,
as shown in the fig. 4, referring to the pre-established windows
of 45,000 sq. km.
The built-up area overcomes the 5,000 sq. km in New York,
Tokyo and Chicago. In London, Sao Paulo and Shanghai is
between 4,000 and 5,000 sq. km. Finally, Moscow and overall
Mexico City has a lower built-up area, 3,000 and 2,000 sq. km
respectively. The above results suggest, if they are correlated
with information on population and economic structure of those
metropolises, a higher level of urban sprawl in the USA and
Japan than in Europe and Latin-speaking America.
