Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
1Activity
0 of .
Results for:
No results containing your search query
P. 1
6_390

6_390

Ratings: (0)|Views: 14 |Likes:

More info:

Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

02/24/2011

pdf

text

original

 
A MODIFIED PARTITION FUSION TECHNIQUE OF MULTIFOCUSIMAGES FOR IMPROVED IMAGE QUALTITY
Dheeraj Agrawal
1
, Dr.Al-Dahoud Ali
2
, Dr.J.Singhai
3
 
1,3
Department of Electronics and Communication Engineering, MANIT, Bhopal. (M.P.), INDIA
2
Faculty of Science and Information Technology, Al-Zaytoolah university of Amman, Jorden.
1
dheerajagrawal@manit.ac.in,
2
aldahoud@alzaytoonah.edu.jo,
3
j_singhai@rediffmail.com
 
ABSTRACT
 This paper presents a modified Partition fusion technique for multifocus images for improvedimage quality. In the conventional partition fusion technique image sub blocks are selected for fused image based on their clearity measures. The clearity measure of an image sub block can bedetermined by second order derivative of the sub image. The performance of these clearitymeasures is insufficient in noisy environment. In the modified technique, before dividing theimage into sub images, it is filtered through linear phase 2-D FIR low pass digital filter toovercome the effect of noise. The modified technique uses choose max selection rule to select theclearer image block from the differently focused source images. Performance of the modifiedtechnique is tested by calculating the value of RMSE. It is found that EOL gives lowest RMSEwith unequal block sizes while SF gives lowest RMSE with equal block sizes when used asclearity measure in modified partition fusion technique.
Keywords:
EOL, RMSE, MI, FIR.
1. INTRODUCTION
The images are the real description of objects. Whenthese images are taken from camera there are somelimitations of a camera system. One of which is thelimitation of depth of focus. Due to this an imagecannot be captured in a way that all of its objects arewell focused. Only the objects of the image with inthe depth of field of camera are focused and theremaining will be blurred. To get an image wellfocused everywhere we need to fuse the images takenfrom the same view point with different focussettings. The term image fusion is used for practicalmethods of merging images from various sensors to provide a composite image which could be used to better identify natural and manmade objects. In therecent research works the researchers have usedvarious techniques for multi-resolution image fusionand multi focus image fusion.
.
Li’ et al.,(2001-2002)introduced a method based on the selection of clearer image blocks from source images[8,9].In thismethod, image is first partitioned into blocks thenfocus measure is used as activity level measurement.Based on activity level, best image block is selected by choosing image block having maximum value of activity for fused image. The advantage of thismethod is that it can avoid the problem of shift-variant, caused by DWT. Also according to theanalysis of the image blocks selection method, theimplementation is computationally simple and can beused in a real-time. The limitation of this method isof its robustness to noise. This method does not perform quit well for noisy images. To overcome thislimitation preprocessing of the image has been donewith the help of a low pass filter.The measure of clarity plays an important role in thiskind of fusion method. A better measure results in asuperior fusion performance. However, little work has been done on the image clarity measures in thefield of multi-focus image fusion. The image claritymeasures, namely focus measures, are deeply studiedin the field of autofocusing. The paper alsoconsidered the fact that the background informationlie in low frequency component of the image; sowhile using different focusing parameters the method proposed will be able to extract the features of  background information when the image is passed bya low pass filter. This paper is organized as follows.A brief description of focus measures is given inSection 2. Proposed modified technique for obtaininglow RMSE fused image is discussed in Sections 3and Sections 4 presents results of the proposedmethod in comparison with existing methods.
2. FOCUS MEASURES
A value which can be used to measure the depth of field from the acquired images can be used as focusmeasure. Depth of field is maximum for the bestfocused image and generally decreases as the defocusincreases.
Special Issue on ICIT 2009 Conference - Bioinformatics and ImageUbiCC Journal Volume 4 No. 3658
 
A typical focus measure satisfies followingrequirements:1. Independent of image content;2. monotonic with respect to blur;3. The focus measure must be unimodal, that is, itmust have one and only one maximum value;4. Large variation in value with respect to the degreeof blurring;5. Minimal computation complexity;6. robust to noise.The conventional focus measures used to measure theclearity of the images are variance, EOG, EOL, andSF. These focus measures are expressed as followingfor an M x N image with f(x, y) be the gray levelintensity of pixel (x, y).
1.
Variance:
 
The simplest focus measure is thevariance of image gray levels. The expression for theM × N image f(x, y) is:
MN211
variance
1(f(,))MN
x y
 x y
= =
=
×
,Where
µ
is the mean value and is given as
MN11
1f(,)MN
x y
 x y
µ 
= =
=
×
∑ ∑
 
2.
Energy of image gradient (EOG):
 
This focusmeasure is computed as:
M1N12211
EOG=
(ff)
 x y x y
= =
+
 
Where
ff(1,)f(,)
 x
x y x y
= +
 
ff(,1)f(,)
 y
x y x y
= +
 3.
Energy of Laplacian of the image (EOL):
 
Itis used
 
for analyzing high spatial frequenciesassociated with image border sharpness is theLaplacian operator.
M1N1222
EOL=
(ff)
 xx yy x y
= =
+
 Where
fff(1,1)4f(1,)f(1,1)4f(,1)20f(,)4f(,1)f(1,1)4f(1,)f(1,1)
 xx yy
x y x y x y x y x y x y x y x y x y
+ =− − − − + + + + + − + +
 
4.
Spatial frequency (SF):
Strictly speakingfrequency is not a focus measure. It is a modifiedversion of the Energy of image gradient (EOG).Spatial frequency is defined as:
22
SF
RFCF
=
+
 Where RF and CF are row and column frequenciesrespectivly:
MN212
RF
1(f(,)f(,1))MN
x y
 x y x y
= =
=
×
∑∑
and
MN221
CF
1(f(,)f(1,))MN
x y
 x y x y
= =
=
×
∑∑
 
5.
Visibility (VI):
This focus measure is inspiredfrom human visual system, and is defined as
MN+1m=1n=1
f(m,n)-VI=
α 
µ µ 
 Where µ is the mean intensity value of the image, and
α
is a visual constant ranging from 0.6 to 0.7.
 
3. MODIFIED TECHNIQUE FOR LOW RMSE
Most of the focus measures are based on theidea of emphasizing high frequency contents of theimage and measure their quantity. This comes froman idea that blurring suppresses high frequenciesregardless of particular Point Spread Function. [13]Considering the performance of various focusmeasures, EOL found to be the best among all [8].Laplacian of an image is determined by second order derivative of the image. The performance of thesecond order derivative decreases if noise is presentin the source images as show in Fig-1Fig-1 the performance of second order derivative in presence of various degree of noise.Fig-1 (A), (E)Fig-1 (B), (F)Fig-1 (C), (G)Fig-1 (D), (H)
Special Issue on ICIT 2009 Conference - Bioinformatics and ImageUbiCC Journal Volume 4 No. 3659
 
Fig-1(A) shows ramp edges profile of an imageseparating black region and white region. The entiretransition from black to white represents a singleedge. In fig-1(A) image is free of noise and its greylevel profile is sharp and smooth.Fig-1(B-D) arecorrupted by additive Gaussian noise with zero meanand standard deviation of 0.1, 1.0 and 10.0 intensitylevels respectively and their respective grey level profile shows noise added on the ramp by rippleeffects. The images in the second column are thesecond derivatives of the images on the left. Fig-1(E)shows two impulses representing presence of edge inthe image.Fig-1(F-H) shows that as the noiseincreases in the image the detection of impulses becomes difficult making it nearly impossible todetect the edge in the image. This shows that thefocus measure using the second order derivative alsofails to decide about the best focused image in noisyenvironment. Thus for selection of best focusedimage removal of noise is essential before applyingfusion technique to obtain best focused image.The proposed focusing technique uses the linear- phase 2-D FIR low pass digital filter to remove thenoise from the differently focused images. Filter usesParks-McClellan algorithm [19], [20].The Parks-McClellan algorithm uses filter with Equiripple or least squares approach over sub-bands of thefrequency range and Chebyshev approximationtheory to design filters with an optimal fit betweenthe desired and actual frequency responses. Thefilters are optimal in the sense that the maximumerror between the desired frequency response and theactual frequency response is minimized. Filtersdesigned this way exhibit an equiripple behavior intheir frequency responses and are sometimes calledequiripple filters. Filters exhibit discontinuities at thehead and tail of its impulse response due to thisequiripple nature.These filters are used in existingfusion algorithm before partitioning the image asshown in fig-3. The source images are passed through2D FIR low pass filter of order 4 and havingcharacteristic as shown in fig-2. For these low passfiltered images conventional focus measure such asVariance, Energy of Gradient, Energy of Laplacian,Spatial frequency are computed.
F
ig.2.
P
erspective plot of linear phase 2-D FIR Lowpass digital filter 
Setup for proposed algorithms
A schematic diagram for proposed image fusionmethod is shown in Fig-3.The paper proposesmodification for obtaining best focus measure innoisy environment by use of filter at step -2 in theexisting algorithms used by Li et. al [8].The fusion method consists of the following steps:
Step 1
. Decompose the differently focused sourceimages into blocks. Denote the ith image block of source images by A
i
and B
i
respectively.
Step 2
. Filter the images through a 2D FIR low passfilter for removal of noise.
Step 3
. Compute the focus measure of each block,and denote the results of A
i
and B
i
by M
iA
and ,M
iB
 respectively.
Step 4
. Compare the focus measure of twocorresponding blocks A
i
, and B
i
and construct the ith block D
i
of the composite image as
>>=
 Ai Bii Bi Aiii
 M  M  B M  M  A D
 Step 5
. Compute root mean square error (RMSE) for the composite image with a reference image
 
Fig.3: Schematic diagram for evaluating proposed focusing technique in Multi-focus image fusion
ActivitylevelmeasureCombining by choose
 max
FIR LPF
 
FIR LPF
A
 
BFused imagePortitionedimages
Special Issue on ICIT 2009 Conference - Bioinformatics and ImageUbiCC Journal Volume 4 No. 3660

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->