experiment 7: image restoration filters (using matlab) by...

8
Experiment 7: Image restoration filters (using MATLAB) By: Dr. Rajeev Srivastava This experiment demonstrates following image restoration filters and their comparative study: 1. Blind Deconvolution algorithm 2. Lucy-Richardson algorithm 3. Regularized filter 4. Weiner filter 1. Blind Deconvolution algorithm It is a deconvolution technique that permits recovery of the target scene from a set of ‘blurred image’ in the presence of a point spread function.It can be performed iteratively, whereby each iteration improves the estimation of the PSF and the scene, or non-iteratively, where one application of the algorithm, based on exterior information, extracts the PSF. Deblurring image using this algorithm includes Reading an image Simulate the blur by using convolving a Gaussian filter with the true image (using imfilter). The Gaussian filter then represents a point-spread function, PSF. Restoring the blurred image using PSF of various size. UNDERSIZED PSF is the array with shorter pixel level,OVERPSF is the array with longer pixel level,INITPSF for an initial PSF that is exactly of the same size as the true PSF. To determine different PSF padarray function is used. Padarray(A,padsize,padval,direction) A is the matrix to b padded padsize :specifies both direction of padding and direction along which to pad. padval : circular->circular repetition of element with in the dimension Replicate->pad by repetition of border element of array. Symmetric->pad with mirror refection of it. Direction: character string specifies direction to apply. The specified directions are: Both: pad before 1 st element n after last element along each direction. Pre: pad before last array element along each direction. Post: pad after last array element along each direction. deconvblind function :

Upload: others

Post on 13-Jul-2020

5 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Experiment 7: Image restoration filters (using MATLAB) By ...content.inflibnet.ac.in/data-server/eacharya... · Original Image Blurred Image Restored Using Initial PSF Fig: Restoration

Experiment 7: Image restoration filters (using MATLAB)

By: Dr. Rajeev Srivastava

This experiment demonstrates following image restoration filters and their comparative

study:

1. Blind Deconvolution algorithm

2. Lucy-Richardson algorithm

3. Regularized filter

4. Weiner filter

1. Blind Deconvolution algorithm

It is a deconvolution technique that permits recovery of the target scene from a set of ‘blurred

image’ in the presence of a point spread function.It can be performed iteratively, whereby

each iteration improves the estimation of the PSF and the scene, or non-iteratively, where one

application of the algorithm, based on exterior information, extracts the PSF.

Deblurring image using this algorithm includes

Reading an image

Simulate the blur by using convolving a Gaussian filter with the true image (using

imfilter). The Gaussian filter then represents a point-spread function, PSF.

Restoring the blurred image using PSF of various size. UNDERSIZED PSF

is the array with shorter pixel level,OVERPSF is the array with longer

pixel level,INITPSF for an initial PSF that is exactly of the same size as the true

PSF.

To determine different PSF padarray function is used.

Padarray(A,padsize,padval,direction)

A is the matrix to b padded

padsize :specifies both direction of padding and direction along which to pad.

padval : circular->circular repetition of element with in the dimension

Replicate->pad by repetition of border element of array.

Symmetric->pad with mirror refection of it.

Direction: character string specifies direction to apply. The specified directions are:

Both: pad before 1st element n after last element along each direction.

Pre: pad before last array element along each direction.

Post: pad after last array element along each direction.

deconvblind function :

Page 2: Experiment 7: Image restoration filters (using MATLAB) By ...content.inflibnet.ac.in/data-server/eacharya... · Original Image Blurred Image Restored Using Initial PSF Fig: Restoration

[J,PSF] = deconvblind(I, INITPSF,NUMIT)deconvolves image I using the maximum

likelihood algorithm, returning both the deblurred image J and a restored point-spread

function PSF.NUMIT is the no of iteration, default is 10.

Analyzing the Restored PSF.

Improving the restoration

Original Image Blurred Image

Restored Using Initial PSF

Fig: Restoration Using Blind Deconvolution Algorithm

2. Lucy-Richardson algorithm

It is an iterative procedure for recovering a latent image that has been blurred by a point

spread function. The algorithm maximizes the likelihood that the resulting image when

convolved with the PSF is an instance of the blurred image, assuming poisons noise statistics.

Maximizing the likelihood function of the model yields an equation that is satisfied when the

following iteration converges.

= ( )

original image

blurred image

deblurring with initial psf

Page 3: Experiment 7: Image restoration filters (using MATLAB) By ...content.inflibnet.ac.in/data-server/eacharya... · Original Image Blurred Image Restored Using Initial PSF Fig: Restoration

‘*’ is the convolution, is the estimate of the undegraded image is the degraded

output image and is the spatial representation of the degraded function.

The approach followed is stopped when a result acceptable in a given application has been

obtained.

deconvlucy(I,psf,nummit) : restore image that was degraded by convolution with point

spread function and an additive noise. Here numit is the no of iteration. Default is 10.

Original Image Blurred Image

Blurred & Noisy Image Restored Using 7 iteration

original Image Blurred

Blurred & Noisy Restored image, iteration=7

Page 4: Experiment 7: Image restoration filters (using MATLAB) By ...content.inflibnet.ac.in/data-server/eacharya... · Original Image Blurred Image Restored Using Initial PSF Fig: Restoration

Restored using 10 iteration Restoration Using 20 Iteration

Fig: Restoration Using Lucy-Richardson Algorithm

3. Regularized filter

It is used effectively when limited information is known about the adiitive noise.

deconvreg() : deconvolves image i using above algorithm returning deblurred image

syntax: [reg1 lagra]=deconvreg(I,psf,NP, parameter) where is the NP is the additive noise

power. reg1 is the restored image and lagra is a scalar,Lagrange multiplier. Over-estimated

NP leads to poor resolution. Under-estimated NP leads to overwhelming noise.

edgetaper(): blurs the edges of input image i using psf.reduce noise amplification and ringing

algorithm d boundary.

syntax: edgetaper(I,psf)

Original Image Blurred Image

Restored image with damping,interation=10

original imageBlurred

Page 5: Experiment 7: Image restoration filters (using MATLAB) By ...content.inflibnet.ac.in/data-server/eacharya... · Original Image Blurred Image Restored Using Initial PSF Fig: Restoration

Edgetar Effect Blurred & Noisy

Restored with Noise Power Restored with Langrange Multiplier

Fig: Restoration Using Regularized Filter

4. Weiner filter: It is one of the best approaches to linear image restoration. It seeks an

estimate that minimizes the statistical error function

} where the expected value operator and is is the undegraded image.

The solution to above expression in the frequency domain is:

| |

| |

Where the degradation function.

| | =

=the complex conjugate of

| | = the power spectrum of noise

| | = power spectrum of undegraded image

⁄ signal to noise power

Edgetaper effect Blurred & noisy

restored with nprestored with lagrange multiplier

Page 6: Experiment 7: Image restoration filters (using MATLAB) By ...content.inflibnet.ac.in/data-server/eacharya... · Original Image Blurred Image Restored Using Initial PSF Fig: Restoration

deconvwnr(I,psf,NSR) : deconvolves image I using weiner filter algorithm. NSR is the noise

to signal ration.

Original Image Blurred Image

Restored Image Simulate Blur & Noise

Restored blurred & noisy image Restored image using estimated SNR

Fig: Restoration Using Wiener Filter

Original Image Blurred Image

Restored ImageSimulate Blur & Noise

Restored blurred & noisy imageRestored blurred,noisy image using estimated snr

Page 7: Experiment 7: Image restoration filters (using MATLAB) By ...content.inflibnet.ac.in/data-server/eacharya... · Original Image Blurred Image Restored Using Initial PSF Fig: Restoration

5. PERFORMANCE ANALYSIS OF DIFFERENT DEBLURRING ALGORITHM

(on example image)

Algorithm MSE RMS PSNR NMSE

Blind Deconvolution

Algorithm

Parameters:-

1. Symmetric

2. Replicate

3. Circular

2.5588e+004

9.2978e+003

7.6797e+003

159.9638

96.4252

87.6341

4.0504

8.4470

9.2773

6.4137e-009

2.3305e-009

1.9249e-009

Lucy-Richardson

Algorithm

Parameters:- Variance=.005,

Iteration=10

1. Symmetric

2. Replicate

3. Circular

4.2047e+004

4.1654e+004

4.0578e+004

205.0525

204.0929

201.4387

1.8935

1.9342

2.0479

1.0539e-008

1.0440e-008

1.0171e-008

Variance=.002,

Iteration=15

1. Symmetric

1. Replicate

2. Circular

1.5623e+004

1.4840e+004

1.4600e+004

124.9988

121.8214

120.8286

6.1924

6.4163

6.4874

3.3938e-008

3.2589e-008

3.2060e-008

Regularized Filter

Parameters:-

1. Symmetric

2. Replicate

3. Circular

6.7710e+004

6.5248e+004

7.8193e+004

260.2120

255.4378

279.6308

-0.1757

-0.0149

-0.8009

1.4068e-007

1.3556e-007

1.6246e-007

Wiener Filter

Parameters:-

Length=11

Theta=21

Noise variance=.0001

1. Symmetric

2. Replicate

3. Circular

0.0719

0.0253

0.0127

0.2682

0.1590

0.1127

59.5623

64.1025

67.0910

1.1722e-009

4.1208e-010

2.0707e-010

Page 8: Experiment 7: Image restoration filters (using MATLAB) By ...content.inflibnet.ac.in/data-server/eacharya... · Original Image Blurred Image Restored Using Initial PSF Fig: Restoration

Noise variance=.002

1. Symmetric

2. Replicate

3. Circular

Noise variance=.02

1. Symmetric

2. Replicate

3. Circular

0.1569

0.1555

0.1512

9.0806

9.0892

9.0863

0.3962

0.3943

0.3888

3.0134

3.0148

3.0144

56.1735

56.2143

56.3352

38.5496

38.5456

38.5469

2.5578e-009

2.5339e-009

2.4644e-009

1.4800e-007

1.4814e-007

1.4809e-007