Sei sulla pagina 1di 6

ENHANCEMENT OF MEDICAL IMAGE FUSION USING JOINT

SPARSE METHOD
MADHUPRIYA B
M.Tech, Dept of ECE, Don BoscoInstitue of Technology, Bangalore, Karnataka, India

S.K.FAIROOZ
Associate Professor, Dept of ECE, Don Bosco Institute of Technology, Bangalore, Karnataka,
India

ABSTRACT: In this paper, the hybrid of Orthogonal Matching Pursuit (OMP), Principal
Component Analysis (PCA) and Discrete Wavelet Transform (DWT) are used, for joint sparse
image fusion method. First we extract common and innovative images from the source images and
the next step is using sparse PCA method for fusion of innovative features. The result of sparse
PCA method is fused with the common features by weighted average fusion method. This helps in
preserving the edge information and high spatial resolution. DWT is applied separately to the
images and inverse DWT of this image is given to the weighted average fusion method. The result
obtained is of better performance in terms of PSNR and MSE.

KEYWORDS: Image fusion, OMP, PCA, DWT, Joint sparse.

INTRODUCTION

Image fusion is done, for the multiple images obtained by using different sensors for the same
scenes. The resultant fused image is more informative, than the single captured image. Medical
images have come far way in the field of image fusion. Various medical images such as, Magnetic
Resonance Imaging (MRI), Positron Emission Tomography(PET), Computerized Tomography
(CT) give different perspective of the same scene, and thus makes the diagnostic process
easier.MRI gives the view of soft tissue and CT gives image of bone structure. This gives scope to
image fusion for the precise diagnosis. The criteriasrequired for fusion of images are : 1) fused
image should preserve all the required information for further processing 2) Artifacts should not be
introduced 3) noise and unnecessary information should be suppressed.

Sparse representation is possible by using different Greedy approaches such as : Matching Pursuit
(MP), Orthogonal Matching Pursuit(OMP) and Stage wise OMP (St OMP). These techniques
represent signals with fewest number of non-zero coefficients.PCA method is powerful approach
in terms of visual inspection and quantitative evaluation metrics. By fusing images of both sparse
and PCA, we overcome the disadvantages of both the methods, that is with the dense non zero
entriesthe uninformative features are represented by PCA method and the sparse method lacks
high spatial resolution.. Advantages of the PCA is, it enhances the spatial quality and the sparse
method advantage is it preserves the important information. Our proposed method overcomes the
disadvantage of sparse fusion and PCA respectively.
METHODS

Orthogonal matching pursuit (OMP) constructs approximations by iteration process. At each


iteration the solution is calculated until it satisfies the conditions.

Pseudo Algorithm of OMP


Given: Dictionary D, signal S, and error threshold
1) Initialize residual , index set I0={} and mainiteration is k = k+1 (initial k=0).
2) Using the ideal solution = , Calculatethe error ( ) = min for all i.
3) Update stage: Augmenting the index set = 1{ 0} (find i0of e( ): 1 and e( 0)
e( ).
4) Update the solution sk( 0)+= and residual.
5) If stopping criterion is met, = ; else, apply anotheriteration.

The sparse codingalgorithms are constructed based on the premise thatDictionary D of size nk is
already known. For effectiveresults, we use phase included DCT (Discrete CosineTransform)
dictionaryfor our experiment. The signal is sparse represented , as given
dictionary .

The Discrete Wavelet transform(DWT) converts the image into frequency domain from spatial
domain. The image is divided into four separate parts, these are HH, HL, LH, and LL. These four
parts represent the frequency areas in the image. This is the first level of decomposition of the
image. The low frequency domain LL is sensitive to the human eyes and the other domains have
more details.

LL1 HL1
DWT
Original
image LH1 HH1

1 level DWT
LL2 HL2 HL1

LH2 HH2

LH1 HH1

2 level DWT
Fig1: Decomposition of image by DWT

Inverse Discrete Wavelet Transform (IDWT) is used for reconstructing the image I(x,y), from the
LL,LH,HL and HH images.

226
PROPOSED METHOD

The proposed system uses PCA transformation, sparse transformation and DWT methods. Firstly,
the common and innovative features are extracted. Next step is different fusion rules are applied to
the coefficients.

The algorithm is as follows:


1) Multiple images in ensemble have one common component and multiple innovative
components {Ii}. Innovative component contain the complementary information
from different images of same scene.
2) of different images of same image i are decomposed into sparse vectors
s1,s2,s3,si via OMP method.
3) Sparse coefficients of innovative components are combined using PCA fusion rules.
The covariance matrix Cs of innovative image is as follows :
Cs= cov (Is)= cov([ s1(:),s2(:)])= (1)
Where, s1,s2: sparse vectors for innovative components
4) Fused PCA image Ip is obtained. The reconstructed image I1 is obtained by dividing
each pixel by number of addition operation performed at each pixel.
5) The weighted average scheme is used to fuse innovative component and common
component.
6) DWT is performed on the input images. These images are fused.
7) Inverse DWT is applied to the fused image and is given to the weighted average
fusion.

Fig2: flow of sparse pcadwt joint fusion algorithm

227
RESULTS

We test the performance of the proposed enhanced image fusion using joint sparse method, and
compare the qualitative and quantitative results. Qualitative measurement is done through the
visual inspection, which include sharpness and noise suppression.

Table 1: performance of fusion methods by quality evaluation metrics


Image Type Fusion PSNR(db) Correlation Mutual Entropy SSIM
strategy Information
Multimodal Joint PCA 26.4111 0.9043 0.9634 6.7409 .9977
(MRI & CT) sparse
PCA 20.8380 0.8690 0.8096 6.5502 .9919
OMP fusion 24.8056 0.8985 0.9440 6.7376 .9975
Joint sparse 33.5582 0.8464 0.8269 5.7045 .3193
and DWT

Fig3: CT input image

Fig 4: MRI input image

228
Fig 5: DWT of fused image

Fig 6: output image

CONCLUSION

Medical image fusion plays an important role in clinical diagnosis. In this paper, we have proposed
enhancement of medical image fusion using joint sparse method using OMP and DWT method.
The experimental results show that the proposed method perform better than PCA and OMP
method.

229
REFERENCES

AnuyogamVenkatraman, JavadAlirezaie, Paul Babyn and AlirezaAhmadian(2014),Medical


Image Fusion Using Joint Sparse Method, Middle East Conference on Biomedical
Engineering.
NishaGawari, Dr. Lalitha .Y .S (2014), Comparative Analysis of PCA,DCT and DWT based
image fusion techniques, International journal of Emerging Research in Management &
Technology.
Philip Breen (2009) , Algorithms for Sparse Apporximation, University of Edinburgh.
M.R. Metwalli ,A.H. Nasr, O.S. Farag Allah and S.El-Rabaie,Image Fusion based on PCA and
High pass filter, Proceedings of the IEEE/ICCES 2009 international conference.
H.Yin, S.Li (2011), Multimodal image fusion with joint sparsity model, Opt Eng,50(6).
Z. Sadeghipour, M. Babaie-Zadeh and C.Jutten, An adaptive thresholding approach for image
denoising using reduntant representations, IEEE international workshop on Machine learning
for Signal Processing.
Zhou Wang, Alan C. Bovik, Hamid R. Sheikh and Eero P. Simoncelli(2004), Image Quality
Assessment: From Error Visibility to Structural Similarity, IEEE transcations on Image
Processing.

230

Potrebbero piacerti anche