Vous êtes sur la page 1sur 35

OPTIMAL INVERSION OF ANSCOMBE TRANSFORM AND FILTER

BASED POISSON NOISE REMOVAL IN MRI DATA SETS


1

Presented by
G.Akshaya Karthika
II Year M.E.,(Communication Systems)

Under the Guidance of


Mr.P.Karthikeyan.
Assistant Professor
Department of ECE

IMAGE DENOISING
2

Noise reduction is the most important for enhancing color images.


The noise arises from improper lighting, movement of objects,

sensitivity of imaging devices, resulting artifacts, blur & contrast


sensitivity.
The image acquisition sensor output carries both signal and noise

components which make high-quality image acquisition difficult.


Removal of Poisson noise is challenging because it is signal

dependent. Its magnitude varies depend upon the intensity of


image, especially it affects the medical images .

Contd.
3

Medical images(X-ray, CT,MRI) are mainly affected by


Poisson noise .Because X-ray distribution follows the
Poisson distribution.

Diagnosing of disease is complicated due to the noise. The


important information is removed if we perform the
denoising techniques.

Points Observed From the Literature


Survey
4

1.V N Prudhvi and Dr T Venkateswarlu,Denoising of Medical


Images using Total Variational MethodSignal & Image
Processing:An International Journal(SIPIJ)Vol.3,No.2 2012

Total Variational Method

Introduces Artifacts

Contd..
5
2.D.Mary Sugantharathnam and Dr.D.Manimegalai,The Curvelet approach for Denoising in
various Imaging modalities Using different shrinkage rulesInternational Journal of Computer
Applications Volume 29-No.7,September 2011.

3.Wavelets, Ridgelets, and Curveletsfor Poisson Noise Removal


Bo Zhang, Jalal M. Fadili, and Jean-Luc StarckIEEE TRANSACTIONS ON IMAGE
PROCESSING, VOL. 17, NO. 7, JULY 2008

Curvelet Approach

Fails to remove the Poisson


noise in medical images.

Contd..
6
4.Mirela Frands,IsabelleE.Magnin (2011) proposed Wavelet thresholding- Based denoising
Method of list mode NLM algorithm for compton imaging, International Journal of Computer
Applications

5.Wavelet-Domain Medical Image Denoising UsingBivariate Laplacian Mixture Model


Hossein Rabbani, Member, IEEE, Reza Nezafat, and Saeed Gazor, Senior Member, IEEE
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 56, NO. 12, DECEMBER 2009

Multiwavelet domain
Only suitable for the x-ray
images.

Contd..
7

6.J.Salmon, C-A.Deledalley, R.Willett, Z.Harmany,Poisson


Noise Reduction with Non-local PCA, Duke Univeresity,ECE
Department Durham,NC,USA,CEREMADE,CNRS-parisDauphine,Paris,France

PCA (Principle Component Analysis)

Dimension reduction is needed

Contd..
8

7. J. Bednar and T.L. Watt (1984), Alpha-trimmed means


and their relationship to median filters, IEEE TransAcoust.,
Speech, Signal Processing, Vol. 32, No.1, pp.
145-153..
Median Filter

Suitable for Salt & Pepper


Noise

Contd..
9

8.Hakan Gray Senel, Richard Alan Peters and Benoit


Dawant. 2002. Topological Median Filter. IEEE Trans on
Image Processing. 11(2):89 -104.

Linear filter

Poor performance on Gaussian


noise

Contd..
10

9.Gaussian Noise Filtering from ECG by Wiener Filter and Ensemble


Empirical Mode Decomposition
Kang-Ming Chang & Shing-Hong LiuReceived: 24 May 2009 / Revised: 13
October 2009 / Accepted: 28 December2009
# 2010 Springer Science+Business Media, LLC. Manufactured in The
United States

Wiener Filtering(Uniform
Filtering throughout The Image)

Blurring of Fine Detail

Contd..
11

10.Optimal Inversion of the Anscombe Transformationin Low-Count


Poisson Image Denoising Markku Mkitalo and Alessandro Foi IEEE
Transactions on image processing, vol. 20, no. 1, January 2011

Anscombe transform
Used to convert Poisson
noise into Gaussian

PROBLEM IDENTIFICATION
12

MRI images are affected by 70% poisson noise, 20% Gaussian noise
and 10% other noises.

So phase II a method for removing both Poisson noise and


Gaussian noise is to be proposed.

OBJECTIVE
13

To Implement an efficient algorithm for removing both


Noises.

Analyzing various filters for removing Gaussian noise .


and choose the best filter for Gaussian noise removal.

FLOW OF WORK
14

1. Anscombe
transform

Poisson Noise Added


Image

2. Gaussian removal
Filter

Converted to Gaussian
Noise

3.Optimal Inversion
Transform

Gaussian Denoised
Output

Desired Denoised Output

POISSON NOISE
15

Poisson noise is defined by

PZi yi yi i e yi Zi !
Z

variance is not uniform .

ANSCOMBE TRANSFORM
16

Step 1:

Stabilize the variance into unitary variance .

Convert the Poisson noise into Gaussian noise.

Anscombe transform is given by

3
f Z 2 Z
8

17

INPUT IMAGE

NOISY IMAGE

Output of Anscombe Transform


18

Gaussian noise Removal Filter


19

Step 2:

Comparing various filters used for Gaussian noise


removal.
o
Gaussian
o
Wiener filter
o
Bilateral
o
BM3D

Choose a suitable filter for Anscombe Transform

GAUSSIAN FILTER
20

Gaussian filter smooths the image by calculating weighted average


in filter box
The weight factor W(x , y)= e a
2
2
Where a is the standard deviation a= x y
2.r 2

DISADVANTAGE

It does not preserve the edges

Wiener filter
21

The Wiener deconvolution method has widespread use in image


deconvolution applications, as the frequency spectrum of most visual
images is fairly well behaved and may be estimated easily.

The frequency Response is given by


G f

H * f S f

H f

DISADVANTAGE:

Introduces bluring effect

2 S f N f

BM3D filter
22

1) finding the image patches similar to a given image patch and grouping them in a
3Dblock
2) 3D linear transform of the 3D block;
3) shrinkage of the transform spectrum coeffcients;
4)inverse 3D transformation
DISADVANTAGE:

For higher noise variance it shows poor performance.

Bilateral filter
23

The bilateral filter is also defined as a weighted average


of nearby pixels,in a manner very similar to Gaussian
convolution.

The difference is that the bilateral filter takes into account

the difference in value with the neighbours to preserve


edges while smoothing.

24

The key idea of the bilateral filter is that for a pixel to

influence another pixel, it should not only occupy a


nearby location but also have a similar value.

1
BF [ I ]
wp

G p q G I
s

Iq Iq

q s

where s , r are spatial parameter and range parameter

Bilateral filter Advantages


25

Its formulation is simple: each pixel is replaced by a weighted


average of its neighbors. This aspect is important because it makes it
easy to acquire intuition about its behavior, to adapt it to applicationspecific requirements, and to implement it.

It depends only on two parameters that indicate the size and


contrast of the features to preserve.

It can be used in a non-iterative manner. This makes the


parameters easy to set since their effect is not cumulative
over several iterations.

Filter output
26

GAUSSIAN FILTER
OUTPUT

WIENER FILTER
OUTPUT

Schedule
27

BM3D FILTER
OUTPUT

BILATERAL FILTER
OUTPUT

Inverse Anscombe Transform


28

Step:3
Inverse transform is given by

E{ f z / y} 2
z 0

3 y z e y
z .
8 Z!

Reconstruction using Inverse Anscombe


Transformation
29

Gaussian Filter

Wiener FIlter

30

BM3D FILTER

BILATERAL FILTER

Performance Analysis
31
GAT+Gaussian

Input Images

GAT+ Wiener

MSE

PSNR

MSE

PSNR

0.34

52.81

0.17

55.78

0.25

54.12

0.06

0.23

54.57

0.51

0.34

GAT+ BM3D

MSE

GAT+Bilateral

PSNR

MSE

PSNR

0.02

66.25

0.01

66.47

60.13

0.01

67.29

0.01

68.03

0.05

61.46

0.01

67.61

0.01

68.48

51.02

0.14

56.68

0.02

65.03

0.02

65.86

52.08

0.10

58.01

0.01

67.43

0.01

67.47

References
32

1.Isabel Rodrigues and Joao Sanches,Denoising of Medical Images


Corrupted by Poisson Noise, IEEE Communication society 2008.
2.D.Mary Sugantharathnam and Dr.D.Manimegalai,The Curvelet
approach for Denoising in various Imaging modalities Using
different shrinkage rulesInternational Journal of Computer
Applications Volume 29-No.7,September 2011.
3.V N Prudhvi and Dr T Venkateswarlu,Denoising of Medical Images
using Total Variational MethodSignal & Image Processing:An
International Journal(SIPIJ)Vol.3,No.2 2012

Contd
33

4.Mary sugantharathnam, manimegalai(2011) proposed the curvelet


approach for denoising in various imaging modalities.
5.V N Prudhvi Raj,Dr T Venkateswarlu(2012) proposed Total
Variational method for denosing the Medical images.

6.Mirela Frands,IsabelleE.Magnin (2011) proposed Wavelet


thresholding- Based denoising Method of list mode MLEM
algorithm for compton imaging
7.Florian Luisier, Member,IEEE,Thierry Blu,senior
member,IEEE,and Micheel Unser,Fellow,IEEEImage
denoising in mixed poisson-gaussian noiseIEEE transactions
on image processing,Vol.20,No.3,March 2011

Contd
34

8. J. Bednar and T.L. Watt (1984), Alpha-trimmed meansand their relationship to


median filters, IEEE TransAcoust., Speech, Signal Processing, Vol. 32, No.1,
pp.145-153..
9. Hakan Gray Senel, Richard Alan Peters and Benoit Dawant. 2002.
Topological Median Filter. IEEE Trans on Image Processing. 11(2):89 104
10. Gaussian Noise Filtering from ECG by Wiener Filter and Ensemble
Empirical Mode Decomposition
Kang-Ming Chang & Shing-Hong LiuReceived: 24 May 2009 / Revised:
13 October 2009 / Accepted: 28 December2009 # 2010 Springer
Science+Business Media, LLC. Manufactured in The United States

35

Vous aimerez peut-être aussi