Site Loader

Abstract: Change detection (CD) is an
extensive research field of remote sensing to analyze the changes between two multispectral/multi-temporal images. It has significant potential within the application of urban
development observance to know how it enlarged within a particular period 1.
There are various standard ways and using a lot of techniques for change
detection. In this paper, we proposed a novel change detection methodology via
Kernel Slow Feature Analysis (KSFA). Before that we are doing image compression
that is one of the preprocessing techniques using Haar Transform. Then, KSFA is
projected to extract the nonlinear temporally invariant options to separate
changes and non-changes, to raise the change chance between corresponding multi-temporal
images. Finally, the multi-temporal images were fused and segmented using
Region Growing Technique.  All the proposed
methodology performed well in image change detection and segmentation.

 

Keywords: Change Detection, Kernel Slow Feature
Analysis (KSFA), Haar Transform, Region Growing Segmentation.

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

I. Introduction:
 Change detection is an important topic
in the field of remote sensing. Remotely sensed data provide a quick
representation of the whole real world. The changes done in earth surface might
be the human or natural disasters like earth quake, flooding and unexpected forest
fire. The human made changes are Infrastructure planning and deforestation. Change
detection helps to monitor all these changes and also helps to maintain the
natural resources 2. Nowadays the high-resolution remote sensing data can
also help us to find and classify the ground details with semantic labels such
as industrial region, commercial region, and residential region. The rest of
this paper is organized as follows: Section II details the proposed change
detection methods. The experiments and discussions are presented in Section
III. Finally, the conclusion is drawn in Section IV.

II.
Methodology:

The
procedure which is followed is explained in figure 1.

The
main steps are:

a)
Pre-processing the multi-temporal images using Haar Transform to enhance the
image by compression.

c)
Implement KSFA algorithm to attain the change probabilities of multitemporal
images.

d)
Change Detection Result.

e)
Appling Region-based segmentation using Region Growing technique.

The block
diagram of the proposed system is shown in figure 1.

      

            
The Fig.1 Block diagram of Change Detection System.

 

a)Haar Transform:

In Change detection, there is a need of
pre-processing and it includes filtering, compression and noise removal. Haar
transform is one of the image compression techniques. The objective of
compression technique is to represent the image pixels with less correlation.  It is a simple, fast, and lossless image
compression technique 3. It’s forward and inverse functions need addition and
subtraction only without convolution. Therefore the computational time and
complexity can be reduced by this transform 4. This compression technique follows
orthogonal function.

The Haar transform yn of an n-input
function xn is

yn = Hnxn 

The inverse Haar transform is

xn  = HT
yn

b) Kernal Slow Feature
Analysis:

Change
detection in multitemporal images is very difficult to find the changes due to
the complex structure. The KSFA is a new method for learning invariant and
slowly varying feature in the fast varying signal 5. It can be applied to
process high-dimensional input signal and extract complex feature. In this
paper, KSFA is applied to classified images to get the pixel variation of
multitemporal images. Based on the difference we can take a binary decision
whether there is a change or non-change between multitemporal images.

SFA
can be explained  by  giving 
a multidimensional temporal signal s(t)=s1(t)….sM(t)T,
Here l ? |t0,t1|
indicate the time, we want to find a set of function g1(s)….,gM(s)
to confirm that the transformed output signals will be temporarily invariant.
That means

min gj ?(gj(s))2?t                             (1)

under the constraints

?gj(s)?t = 0 zero mean              (2)

?(gj(s))2?t =1 unit variance       (3)

?i (T) Pixels are not
homogeneous.

This way the input images were portioned.
The selection of seed point and threshold value is important here to get good partitioning.
This approach will not work for non-smoothly varying regions. Ex: texture.

Post Author: admin

x

Hi!
I'm Erica!

Would you like to get a custom essay? How about receiving a customized one?

Check it out