Fully Automatic Road Network Extraction From Satellite Images
#1

Fully Automatic Road Network Extraction From Satellite Images
Sameeha S & Sreedevi D V
S8, Applied Electronics and Instrumentation
LBS Institute of Technology for Women, Poojapura, TVM



Abstract
Our paper deals with the automatic detection of roads in satellite images. Suggested approach comprises of
preprocessing the satellite image via a series of wavelet based filter banks based on frequency response of the
corresponding FIR filter. Here we use a Trous algorithm twice with two different wavelet bases in order to filter
and denoise the satellite image. The resulting two images are fused together in to a single image of same size as the
original satellite image using KLT transform which is based on principal component analysis
(PCA). Then a fuzzy inference algorithm is used to detect roads based on statistical information and on geometry
which classifies each pixel as road or non-road with regard to fuzzy inference rules yielding in a binary image. This
output is then fed as input to the geographical information system (GIS) for cartographic or for other purpose that
are in need.

Introduction
Our paper deals with a fully automatic road detection
algorithm. Road detection algorithms can be
classified into two major groups; semiautomatic and
automatic. The first approach necessitates that the
user specify some initial conditions usually in the
form of seed points entered manually by a human
operator thorough some graphical user interface
(GUI). The other approach which is fully automatic
on the other hand does not require input from an
operator and works on its own. Here a fully
automatic approach is implemented.
Suggested approach comprises of preprocessing the
image via a series of wavelet based filter banks and
reducing the yielding data into a single image which
is of the same size as the original satellite image, then
utilizing a fuzzy inference algorithm to carry out the
road detection whose output can then be used as an
input to a geographical information system (GIS) for
cartographic or other purposes for that matter. We
use “a trous” algorithm twice with two different
wavelet bases in order to filter and de-noise the
satellite image. Each wavelet function resolves
features at a different resolution level associated with
the frequency response of the corresponding FIR
(finite impulse response) filter. Resulting two images
are fused together using Karhounen- Louve transform
(KLT). This process underlines the prominent
features of the original image as well as denoising it,
since the prominent features appear in both of the
wavelet transformed images while noise does not
correlate well between high and low resolution scales
as it lacks coherence.
On this image obtained thorough wavelet filtering
and KLT, road detection is carried out using a
fuzzylogic inference algorithm. Linguistic variables
used for this task are mean, standard deviation which are computed within a 5x5 pixel size image window
and also another linguistic variable based on
geometry. The inference 2 algorithm then classifies
each pixel as road or non-road with regard to the
fuzzy inference rules yielding in a binary image.
Besides road detection fuzzy logic is a powerful and
intuitive (i.e. in the sense of human friendliness) tool
for identification of other features like runways,
moving and/or stationary targets, other man-made
objects on earth or any combination of these thereof,
etc. Parallelizability of any detection algorithm as a
direct consequence is of utmost importance should it
be intended for military use as well.
The images to be used have 512 x 512 pixel
resolutions. The spatial resolution on earth surface is
about 1.5 m/pixel for each image with only slight
variation from one another.

Wavelet Filtering
Since satellite images are finite energy (i.e. square
integrable) functions a wavelet transform exists.
Numerical algorithm utilized to deconvolve the
normalized temperature images is referred as
“algorithme à trous. It is a translation invariant form
of DWT (Discrete Wavelet Transform) since it does
not involve any decimation during down-sampling.
Because of its good time frequency localization
characteristics wavelet analysis find wide
applications. Wavelet transform decomposes a signal
into a set of basis functions. These basis functions are
called wavelets. Wavelets are obtained from a single
prototype wavelet called mother wavelet by dilations
and shifting. The wavelet transform is computed
separately for different segments of the time-domain
signal at different frequencies. It is designed to give
good time resolution and poor frequency resolution at
high frequencies and good frequency resolution and
poor time resolution at low frequencies. Here we use
Trous Algorithm. The wavelet function is given by
two functions that is, a scaling function and wavelet
function, which represents a low pass filter and high
pass filter respectively. The wavelet function used for
implementing wavelet filtering is:
- db1 (Haar wavelet function)
- db8 wavelet function.

Haar Wavelet function
The Haar function is also known as the db1 wavelet.
Haar system is the unique one that satisfies bi-
orthogonality, symmetry. Haar system has a
symmetric scaling function, an antisymmetric
wavelet function, a single vanishing moment and has
finite support length in the interval [0,1]

Db8 wavelet function
Db8 is daubechies wavelet with more number of
coefficients than Haar wavelets. As the number of
coefficient increases the wavelet becomes smoother.

Wavelet Filtering
Algorithm

The image is fed to a low pass as well as high pass
wavelet filter. The image represented as a matrix is
convoluted by the FIR coefficients of this filter. First
the row wise convolution occurs where the number of
columns is halved. This occurs for both low pass
filter and high pass filter. Thus row wise filtered each
output from both filters are again fed to the scaling
filter and wavelet filter for column wise filtering.
Thus there will be four output images each with size
1/4th of the original image. This can be continued for
any number of levels. The number of level depends
on the resolution required. Here the problem with an
ordinary algorithm is that the size of the image is
decreased which necessitate the decimation during
down sampling. This limitation is overcome by Trous
algorithm.
Trous Algorithm
Trous Algorithm is similar to fast biorthogonal
wavelet transform without subsampling.
For any filter h[n], hj[n] is the filters obtained by
inserting 2j-1 zeros between each sample of
h[n].Inserting zeros in the filters creates holes (trous
in French).Now filtering is performed in a similar
way over these filters with new coefficients.

Why Trous Algorithm?
1. No need for down sampling
2. Obeys Linear Additive reconstruction:
Reconstruction of the original image by adding

the detail coefficients of each level to the smoothened
image
3. Ease of interpretation If the smoothing operation is
stopped at resolution p, reconstruction of the original
image I is achieved by adding the detail coefficients
wj of each level to the smoothed image cp. This
linear simple additive reconstruction formula is the
unique convenience of AWT along with its
translation invariance and both account for its ease of
interpretation. As a consequence of these two
properties this algorithm is often employed in object
detection. Only drawback of this algorithm is its
redundancy, which requires (p+1) times larger
storage space than the original image.
Above convolution can be cast into the following
discrete form. j denotes scale as emphasized earlier,
x1 ,x2 are the pixel coordinates and k1,k2 are dummy
integers which the summations are made onto and g
is the corresponding FIR filter associated with the
scaling function _.
4. Here a Trous algorithm is truncated at p=4 level
which gives adequate information at that point. The
wavelet planes shown in are accumulation of
w1….w4. Note that due to the additive reconstruction
property of the trous algorithm this is equivalent to I-
c4. Therefore in implementation we continue with the
computation of the coarse scale image using until
then subtract it from the original image.
For haar, output image :
I1= I – c’4 for h,g _ db2
For db8, output image:
I2= I – c’’4 for h,g _ db8

Karhounen-Louvetransform
(Klt)

Karhounen –Louve Transform (KLT) is used for
image fusion It tend to decorrelate the components of
a given signal After the wavelet transforms we have
two images, let us call them I1and I2. Then we can
proceed to define matrix X which has the information
about these two images in its two rows .Operator
“vec” denotes vectorization of matrices.
X= vec (I1)
vec (I2) 2xN
2
Given X next step is to compute the Karhounen-
Louve transform of X and collapse the information
into half along its principal components.
Next calculating the mean column vector as in
mx = _ xi / 2= [m1……..m512 2]

Reply

Important Note..!

If you are not satisfied with above reply ,..Please

ASK HERE

So that we will collect data for you and will made reply to the request....OR try below "QUICK REPLY" box to add a reply to this page
Popular Searches: algorithms for quadtree to compress satellite images, ppt on fully automatic emergency light, fully automatic pneumatic system, automatic road extraction using high resolustion satellite imges ppt, ppt extraction text images, road extraction from satellite images project applications, fully automatic digital instruments in wikipedia,

[-]
Quick Reply
Message
Type your reply to this message here.

Image Verification
Please enter the text contained within the image into the text box below it. This process is used to prevent automated spam bots.
Image Verification
(case insensitive)

Possibly Related Threads...
Thread Author Replies Views Last Post
  information about satellite based project 1 2,152 09-03-2017, 12:26 PM
Last Post: jaseela123d
  Wireless Power Transmission via Solar Power Satellite full report project topics 32 50,216 30-03-2016, 03:27 PM
Last Post: dhanabhagya
  A Solar Panels Automatic Tracking System Based on PLC seminar class 1 3,493 22-03-2015, 11:25 PM
Last Post: [email protected]
  ARTIFICIAL NEURAL NETWORK AND FUZZY LOGIC BASED POWER SYSTEM STABILIZER project topics 4 6,139 28-02-2014, 04:00 AM
Last Post: Guest
  Authentication Schemes for Session Passwords using Color and Images computer girl 2 2,356 20-09-2013, 10:52 AM
Last Post: computer topic
  MICROCONTROLLER BASED AUTOMATIC POWER FACTOR CONTROLLING SYSTEM projectsofme 2 5,590 20-07-2013, 10:39 AM
Last Post: Mitesh Diwakar
  SOLAR AUTOMATIC MOBILE CHARGER WITH PAY SYSTEM seminar class 13 11,128 12-07-2013, 11:28 AM
Last Post: computer topic
  Low Power Wireless Sensor Network computer science crazy 4 5,737 30-04-2013, 10:04 AM
Last Post: computer topic
  AUTOMATIC DOOR In theatres WITH DOWN COUNTER DISPLAY seminar class 3 2,811 04-04-2013, 10:04 PM
Last Post: Guest
  A Disaster Information System by Ballooned Wireless Adhoc Network seminar surveyer 2 2,365 15-02-2013, 10:20 AM
Last Post: seminar details

Forum Jump: