You are on page 1of 33

NARAYANA ENGINEERING COLLEGE:: GUDUR

(Autonomous)
(Recognized by UGC 2(f) and 12(B), An ISO 9001-2015 certified Institution, Approved by AICTE New Delhi, Affiliated Permanently to
JNTUA, Anantapuramu, Accredited with “A” Grade by NAAC) Gudur, SPSR Nellore (DT), A.P-524101

Department of Electronics and Communication Engineering

A Novel Image Fusion Technique using Multi-Resolution Singular Value


Decomposition
Under the esteemed guidance Of
Mrs. Y. NEERAJA , MTech, (Ph. D)
Associate Professor, Department of E.C.E
BATCH NO B1 Presented by :
S. Rupa (17F11A04A1)
Sk. Haseena (18F15A0411)
T. Srinivasulu (17F11A04A8)
V. Ashwin (17F11A04B4)
M. Ajay (16F11A0453)

1
Contents
Abstract
Fusion Introduction
Design metrics
Fusion using DCT based Laplacian pyramid
Fusion using wavelet transform
Fusion using MSVD
Comparison between three techniques
Conclusions

2
ABSTRACT
 Image fusion is the process of extracting meaningful visual
information from two or more images and combining them
to form one fused image.

 In this project, Laplacian pyramid and wavelet based


fusion schemes are implemented and MSVD based fusion
technique is proposed.

 Simulation results prove that the MSVD scheme


outperform other schemes.
3
Fusion Introduction
 Image fusion is the process that combines information
from multiple images of the same scene.

 The result of image fusion is a new image that retains


the most desirable information and characteristics of
each input image

 The main application of image fusion is merging the


gray-level high-resolution panchromatic image and
the colored low-resolution multispectral image.
4
Design metrics
Root mean square error:

Peak signal to Noise ratio:

5
Fusion using DCT based Laplacian pyramid
Image fusion was done by using Laplacian Pyramid.

First, it constructs the Laplacian pyramid of the source images.

Second, it does the fusion at each level of the decomposition.

Finally, it reconstructs the fused image from the fused pyramid.

Subjective tests are performed to analyze the implemented


techniques.

6
Laplacian Results

figure1 figure2

7
Figure4:Error
Figure3:output
8
Image fusion using wavelet transform
Image fusion scheme, the source images I1 (x, y)
and I2 (x, y) , are decomposed into approximation
and detailed coefficients at required level using
DWT.
The approximation and detailed coefficients
of both images are combined using fusion rule.
The fused image If ( x, y ) could be obtained by taking
the inverse discrete wavelet transform
(IDWT)

9
Wavelet Results

figure1 figure2
10
Figure3:output Figure4:Error
11
PROPOSED METHOD

 SVD
 MSVD Fusion Process
 Simulation Results
 Performance Analysis

12
SVD
Singular Value Decomposition

A factorization of a given matrix to its components:

M = UΣV∗

When:
M – an m x n real or complex matrix
U – an m x m unitary matrix, called the left singular vectors
V – an n x n unitary matrix, called the right singular vectors
Σ – an m x n rectangular diagonal matrix, called the singular values
Applications and Intuition
If M is a real, square matrix –
U,V can be referred to as rotation matrices and Σ as a
scaling matrix
M = UΣV∗
Applications and Intuition
The columns of U and V are orthonormal bases

Singular vectors (of a square matrix) can be interpreted as the


semi-axes of an ellipsoid in n-dimensional space

SVD can be used to solve homogeneous linear equations


Ax=0, A is a square matrix  x is the right singular vector which
corresponds to a singular value of A which is zero

Low rank matrix approximation


Take Σ of M and leave only the r largest singular values, rebuild the
matrix using U,V and you’ll get a low rank approximation of M
Fusion using MSVD
 Wavelets transform, where signal is filtered separately by low
pass and high pass finite impulse response (FIR) filters and the
output of each filter is decimated by a factor of two to achieve
first level of decomposition

 The decimated low pass filtered output is filtered separately by


low pass and high pass filter followed by decimation by a factor
of two provides second level of decomposition.

16
MSVD algorithm
Main Function:
 
Step 1: IM  Read reference image.
IM1 Read the first image.
IM2 Read the second image.
Step 2: -Apply the two input images to the fusion function which gives the

resultant image.
Step3: [X1, U1]  MSVD(IM1)
Step 4: [X2, U2]  MSVD(IM2)

17
MSVD algorithm
Main Function:
Step 5: Prepare LL, LH, HL and HH components (of an image say X) from
the corresponding parts of the images X1 and X2 by using the
following rule.
i) For LL component take average of that of X1 and X2.
ii) For the remaining components take from X1 or X2 whichever is high.
Step 6: U  ½ (U1 + U2)
Step 7: imf IMSVD(X, U)
Step 8: Calculate RMSE and PSNR between the reference and resulting

image.

18
MSVD Function:
Input: Image – x
Outputs : MSVD coefficients – Y, Unitary matrix (U in SVD)
Step 1: m, n  size(x)/2
Step 2: A Zero matrix of order 4xm*n
Step 3: A  x (reshape x into the format of x)
Step 4: [U,S] svd(x)

19
MSVD Function:
Step 5: T  U*A
Step 6: Y.LL  First row of T (reshaped into mxn matrix)
Y.LH  Second row of T (reshaped into mxn matrix)
Y.HL  Third row of T (reshaped into mxn matrix)
Y.HH  Fourth row of T (reshaped into mxn matrix)

20
IMSVD Function:
Inputs: MSVD coefficients – Y, Unitary matrix (U in SVD)
Output: Fused Image – X
Step 1: m, n  size(Y.LL)
Step 2: mn m*n
Step 3: T  Zero matrix of order 4xm*n
Step 4: T  Y (each of four components as rows, so that T is a matrix
of order 4xm*n)
Step 5: A  U*T
Step 6: X  Zero matrix of order 2mx2n
Step 7: X  A (by reshape)

21
CODE
close all;
clear all;

% reference image (ground truth)


imt=double(imread('saras5t.jpg'));

%images to be fused
im1 = double(imread('saras51.jpg'));
im2 = double(imread('saras52.jpg'));

%display images to be fused


figure(1); imshow(im1,[]);
figure(2); imshow(im2,[]);

%apply MSVD
[X1, U1] = MSVD(im1);
[X2, U2] = MSVD(im2);

22
%fusion starts
X.LL = 0.5*(X1.LL+X2.LL);
D = (abs(X1.LH)-abs(X2.LH)) >= 0;
X.LH = D.*X1.LH + (~D).*X2.LH;
D = (abs(X1.HL)-abs(X2.HL)) >= 0;
X.HL = D.*X1.HL + (~D).*X2.HL;
D = (abs(X1.HH)-abs(X2.HH)) >= 0;
X.HH = D.*X1.HH + (~D).*X2.HH;

%XX = [X.LL, X.LH; X.HL, X.HH];


U = 0.5*(U1+U2);

%apply IMSVD
imf = IMSVD(X,U);

%display fused image


figure(3); imshow(imf,[]);title('Fused Image');

%error image
imd = imt-imf;
figure(4); imshow(imd,[]);title('Error Image');
23
Simulation Results

figure1 figure2

24
Figure4:Error
Figure3:output
25
Performance Analysis

Laplacian Wavelet MSVD

26
Performance Analysis
SI.NO METHODS PSNR RMSE

1 LAPLACIAN 38.54 9.16


PYRAMID

2 WAVELET 44.43 2.36


TRANSFORM

3 MSVD 47.63 1.13

27
Applications
• Surveillance
• Military
• Remote sensing
• Machine vision
• Computer vision
• Robotics
• Medical imaging etc.

28
1.Remote Sensing :
In the remote sensing field , the color information is provided
by three sensors covering the red, green and blue spectral
wavelengths.
These sensors have a low number of pixels(low spatial
resolution) and the small objects and details are hidden.
The hidden objects can be observed with different sensor which
have higher number of pixels (high spatial resolution) but no
colour information.
With the fusion process a unique image can be achieved
containing both : High spatial resolution and colour
information .

29
2.Satellite imaging:
In satellite imaging ,two types of images are available.
The panchromatic image acquired by satellites is
transmitted with the maximum resolution available and
the multispectral data are transmitted with coarser
resolution. This will usually be two or four times lower.
At the Receiver station, the panchromatic image is merged
with the multispectral data to convey more information.

30
Future Scope
The proposed scheme of image fusion may be extended
to the following applications.
Video fusion
Background words removal on a written documents.
Graphic designing.

31
Conclusion
In this project, fusion was conceived using different
techniques.

Laplacian and wavelet based fusion schemes are


implemented and a novel MSVD based fusion scheme was
proposed.

From the above comparison table we can clearly say that


MSVD gives the better results.

32
K U
HA N
T

You might also like