A PROPOSED MODEL FOR SEGMENTATION OF SPOT IMAGES
Jalal amini
Department of Surveying Engineering, Faculty of engineering, University of Tehran, Tehran, Iran.
jamini@ut.ac.ir
KEY WORDS: Photogrammetry, Automation, Extraction, Fuzzy, Image
ABSTRACT:
Images are the most important data sources that used for problems in photogrammetry such as automatic object extraction, automatic
orientation and so on. In the most problems, it needs to segment an image to unknown regions for further analysis. In this paper
applied a model for segmentation of an image. The model is applied base on smoothed histogram of the image. First, the histogram
of the image is smoothed with gaussian kernel and then, the regions with similar properties are detected based on the automatic
extraction of peak and valley points in the smoothed histogram to segmentation of image. In the model, we consider the common
areas between the objects as fuzzy areas, The model is tested on the Spot images from Iran.
l. INTRODUCTION
Automatic information extraction of the terrain surface in the
fields of photogrammetry and remote sensing requires the
formulation of procedures and knowledge that encapsulate the
content of the images. This is a non-trivial task, because of the
complexity of the information contained in the images, Images
of the terrain surface in photogrammetry may have scales
varying from 1:3000 to a smaller scale of 1:90000 while in
remote sensing the pixel footprints usually vary from 1 to 30m
(Sowmya and Trinder, 2000). The structure of the features in
images of the terrain is complex, being a combination of many
different intensities that can represent natural features, man
made objects such as buildings and roads , shadows and other
changes in brightness. In addition , when the scale increased,
the content in which features occur becomes considerably more
complex. These characteristics mean that extraction of
information in aerial and satellite images presents major
challenges.
In order to use automatic mapping or special data
acquisition and updating, identification of cartographic objects
in aerial/satellite images has attracted increasing attention in
recent years in digital photogrammetry and remote sensing.
Extraction of roads as man-made objects are attended in
automatic mapping. Therefore, road detection is necessary for
the automation of extracting roads from images. In recent years
a number of papers are published in road extraction
(Baumgartner et al, 1999: Barzohar and Cooper, 1993;
Couloigner and Ranchin, 2000; Gruen, 1997). Some papers
applied semi-automatic methods and others used automatic
methods. When we use automatic methods, it is necessary to
make clear roads from images.
The high capability of high resolution satellite imagery (HRSI),
that includes spatial, spectral, temporal, radiometric resolution
and stereoscopic capability is a powerful new source for the
photogrammetry and remote sensing. Therefore there is a wide
desire to extracting accurate 2D and 3D terrain information
from Im satellite imagery. Also, there are four advantages of
high resolution satellites (Li, 1998). First, the highest resolution
ever available to the civilian mapping community. Second,
extremely long camera focal length for capturing terrain relief
information from satellite orbit. Then, for-nadir and aft-looking
linear CCD arrays supplying in-track stereo strips and “
pointing capabilities generation cross-track sterco strips.
Finally, a base-height ratio of 0.6 and greater.
In this paper, we use a method based of fuzzy sets for
segmentation of Spot images.
2. FUZZY MODEL
Recently, fuzzy set theory has been widely used in
photogrammetry and remote sensing. Researchers like Wang
(1989) applied fuzzy set theory in an expert system for remote
sensing image analysis, Moon (1991) used fuzzy logic system
for integration of different data sets and Krishna Mohan et al.
(2000) used fuzzy logic approaches for land use classification
on IRS-1A L2 data. Here, we used fuzzy logic system for road
identification from IKONOS images.
The complete structure of a fuzzy logic system for
segmentation consist of three steps: First, all input, real
variables (Means and SDs), have to be translated into linguistic
variables, Mean and SD, (fuzzification step). Second step called
fuzzy inference that evaluates the set of IF-THEN rules that
define system behavior. The last step called defuzzification that
translates the fuzzy inference results into a real values as
output. Here, we used three linguistic variables with their terms
as follows:
Mean = {Down-bad, Down-prob-good, Good. Up-prob-
good, Up-bad)],
Standard-Deviation(SD) = (Down-bad, Down-prob-
good, Good, Up-prob-good,
Up-bad], and
Gray-Scale(GS) = fDown-bad, Down-prob-road, Road,
Up-prob-road, Up-bad].
Two membership functions: II shape and Gaussian shape,
are used for all the variables terms. We defined vector X as
input linguistic variables (Mean and SD), and Y which include
the output state linguistic variable, Gray-Scale, For every
linguistic variable, each term is defined by its membership
International
function
Hy) = WU
In the fuzzy |
situation an
variable. Th
two componc
1) Computat
of the THEN
IFM
Scale = Roac
At the end c
linguistic va
translated ir
between lin;
always givet
paper, the
defuzzificati
with membe
output is:
Ls
= eu
y S.
J
The st
[ran in multi
In last sectic
II shape :
segmentatio:
for input anc
