on comparative study of image restoration algorithms

37

Upload: faizan-ali

Post on 19-Jul-2015

32 views

Category:

Science


3 download

TRANSCRIPT

Page 1: On Comparative study of Image Restoration Algorithms

Comparative Study of Image Restoration Algorithm

Faizan Abbas Ali (EE10B081)Under the guidance of Dr. Aravind R.

Department of Electrical Engineering

IIT Madras

May 15, 2015

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 2: On Comparative study of Image Restoration Algorithms

Outline

� Introduction

� Motivation� State of Art

� Image Restoration Techniques

� Optimization Methods� Wiener Filtering� Bayesian Image Restoration

� Results and Discussion

� Convergence of Richardson-Lucy algorithm� C++ Implementation of Richardson-Lucy algorithm

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 3: On Comparative study of Image Restoration Algorithms

Outline

� Introduction� Motivation� State of Art

� Image Restoration Techniques

� Optimization Methods� Wiener Filtering� Bayesian Image Restoration

� Results and Discussion

� Convergence of Richardson-Lucy algorithm� C++ Implementation of Richardson-Lucy algorithm

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 4: On Comparative study of Image Restoration Algorithms

Outline

� Introduction� Motivation� State of Art

� Image Restoration Techniques� Optimization Methods� Wiener Filtering� Bayesian Image Restoration

� Results and Discussion

� Convergence of Richardson-Lucy algorithm� C++ Implementation of Richardson-Lucy algorithm

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 5: On Comparative study of Image Restoration Algorithms

Outline

� Introduction� Motivation� State of Art

� Image Restoration Techniques� Optimization Methods� Wiener Filtering� Bayesian Image Restoration

� Results and Discussion� Convergence of Richardson-Lucy algorithm� C++ Implementation of Richardson-Lucy algorithm

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 6: On Comparative study of Image Restoration Algorithms

Introduction

What is Image Restoration?

� Recovering true image from degraded image

Scope and why is it Interesting?� Surveillance

� Out-of-Focus Images

� Remote Sensing

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 7: On Comparative study of Image Restoration Algorithms

Introduction

What is Image Restoration?

� Recovering true image from degraded image

Scope and why is it Interesting?� Surveillance

� Out-of-Focus Images

� Remote Sensing

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 8: On Comparative study of Image Restoration Algorithms

Introduction

What is Image Restoration?

� Recovering true image from degraded image

Scope and why is it Interesting?� Surveillance

� Out-of-Focus Images

� Remote Sensing

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 9: On Comparative study of Image Restoration Algorithms

Introduction

What is Image Restoration?

� Recovering true image from degraded image

Scope and why is it Interesting?� Surveillance

� Out-of-Focus Images

� Remote Sensing

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 10: On Comparative study of Image Restoration Algorithms

Introduction

State-of-Art

� Classi�cation� non-blind de-convolution� blind de-convolution

� Assumptions

� Linear shift invariant blur degradation model

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 11: On Comparative study of Image Restoration Algorithms

Introduction

State-of-Art� Classi�cation

� non-blind de-convolution� blind de-convolution

� Assumptions

� Linear shift invariant blur degradation model

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 12: On Comparative study of Image Restoration Algorithms

Introduction

State-of-Art� Classi�cation

� non-blind de-convolution� blind de-convolution

� Assumptions� Linear shift invariant blur degradation model

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 13: On Comparative study of Image Restoration Algorithms

Introduction

State-of-Art� Classi�cation

� non-blind de-convolution� blind de-convolution

� Assumptions� Linear shift invariant blur degradation model

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 14: On Comparative study of Image Restoration Algorithms

Optimization Algorithm

Least Square Estimation

f � argminf ||g�Hf||2 � H�g

Steepest Descent and Conjugate Gradient Algorithms

� f pi�1qpx , yq �f piqpx , yq � αhT px , yq � pgpx , yq � hpx , yq � f piqpx , yqq

� Conjugate gradient searches for minima in opposite directionof gradient

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 15: On Comparative study of Image Restoration Algorithms

Optimization Algorithm

Least Square Estimation

f � argminf ||g�Hf||2 � H�g

Steepest Descent and Conjugate Gradient Algorithms

� f pi�1qpx , yq �f piqpx , yq � αhT px , yq � pgpx , yq � hpx , yq � f piqpx , yqq

� Conjugate gradient searches for minima in opposite directionof gradient

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 16: On Comparative study of Image Restoration Algorithms

Optimization Algorithm

Least Square Estimation

f � argminf ||g�Hf||2 � H�g

Steepest Descent and Conjugate Gradient Algorithms

� f pi�1qpx , yq �f piqpx , yq � αhT px , yq � pgpx , yq � hpx , yq � f piqpx , yqq

� Conjugate gradient searches for minima in opposite directionof gradient

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 17: On Comparative study of Image Restoration Algorithms

Image Restoration

Wiener Filtering

W pu, vq � H�pu,vq

|Hpu,vq|2�Snpu,vqSf pu,vq

Pseudo-codefor k Ñ 0 to 2 do

calculate H � H�

H2�k

calculate estimate F � HG

�nd F such that

b|F � F |2 is minimum

output Fend for

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 18: On Comparative study of Image Restoration Algorithms

Image Restoration

Wiener Filtering

W pu, vq � H�pu,vq

|Hpu,vq|2�Snpu,vqSf pu,vq

Pseudo-codefor k Ñ 0 to 2 do

calculate H � H�

H2�k

calculate estimate F � HG

�nd F such that

b|F � F |2 is minimum

output Fend for

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 19: On Comparative study of Image Restoration Algorithms

Image Restoration

Wiener Filtering

W pu, vq � H�pu,vq

|Hpu,vq|2�Snpu,vqSf pu,vq

Pseudo-codefor k Ñ 0 to 2 do

calculate H � H�

H2�k

calculate estimate F � HG

�nd F such that

b|F � F |2 is minimum

output Fend for

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 20: On Comparative study of Image Restoration Algorithms

Image Restoration

Wiener Filtering of SIV gaussian blurred images with added WGN

σb � 0.5 σn � 15 σb � 1.2 σn � 15

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 21: On Comparative study of Image Restoration Algorithms

Bayesian Image Restoration - Richardson-Lucy Algorithm

Richardson-Lucy Algorithm

f pi�1qpx , yq ��

gpx ,yqhpx ,yq�f piqpx ,yq

� hT px , yq�� f piqpx , yq

Pseudo-code

degraded image g and kernel kinitial estimate f0 � g kinput number of iteration nfor iteration 1 to n do

Subtract from blurred image b � g � fi kAdd the error corrected image a � b � fiSet image for next iteration fi � a

end for

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 22: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

Explanation of deconvlucy MATLAB

� Parameters� NUMIT� READOUT� SUBSMPL

� Preprocessing of inputs

� L-R iterations

� Error correction matrix and positivity constraints� Convolving error corrected matrix with blur kernel� Calling corelucy� Setting next -> previous

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 23: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

Explanation of deconvlucy MATLAB contd...

� corelucy

� Convolving input with the blur kernel� Reshape on number of non-singleton dimensions� Estimating image for next iteration

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 24: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

Convergence of RL algorithm

Figure: MSE error of an estimated

image and original image vs log of

iteration of RL algorithm performed

� grayscale images were used

� SIV blurring with addedWGN

� non-blind deconvolutionperformed

� Algorithm is diverging

� deconvlucy MATLAB and ourimplementation performedidentically

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 25: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

Convergence of RL algorithm contd...

� Comparison of performance of RL algorithm

� Grayscale images SIV blurred and noisy

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 26: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

Convergence of RL algorithm contd...

Comparison of deconvlucy for blind and non-blind deconvolution

� Only blurred images converges at around 10 iterations

� Divergent behavior of RL algorithm due to noise

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 27: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

Convergence of RL algorithm contd...

Comparison of deconvlucy for blind and non-blind deconvolution

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 28: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

Convergence of RL algorithm contd...

Comparison of deconvlucy and our implementation for blind deconvolution

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 29: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

Convergence of RL algorithm contd...

Comparison of deconvlucy and our implementationfor blind and non-blind deconvolution

using SSIM index

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 30: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

C++ implementation of RL algorithm

� opencv image processing modules used

� IplImage matrices used

� Cross-platform compatibility code

� opencv functions used

� cvFilter2d� cvCreateImage� cvSub and cvAdd

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 31: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

C++ implementation of RL algorithm

� opencv image processing modules used

� IplImage matrices used

� Cross-platform compatibility code

� opencv functions used� cvFilter2d� cvCreateImage� cvSub and cvAdd

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 32: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

C++ implementation of RL algorithm

True Image

C++ implementation output

Motion Blurred Image

deconvlucy outputFaizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 33: On Comparative study of Image Restoration Algorithms

Richardson-Lucy Algorithm

C++ implementation of RL algorithm

True Image

C++ implementation output

Blurred Image

deconvlucy output ]Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 34: On Comparative study of Image Restoration Algorithms

C++ Implementation of Super-resolution

� Working output from MATLAB C coder not obtained

� opencv image processing modules used

� Modules implemented� header.h� params.h� imLapband.cpp� imagePatches.cpp� DataGenerate.cpp� ImageLumChrom.cpp� test2Patches.cpp� superResolvingAlgo.cpp

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 35: On Comparative study of Image Restoration Algorithms

C++ Implementation of Super-resolution

� opencv functions used� cvtColor� split� merge� GaussianBlur� clone� add� subtract� size� Filter2d

� Implementation issue faced� cvMat data type was used� meshgrid was implemented using nested for loops

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 36: On Comparative study of Image Restoration Algorithms

Conclusion

� blind and non-blind image restoration algorithms werereviewed

� Richardson-Lucy algorithm was studied in-depth

� Convergence of Richardson-Lucy algorithm was analyzed

� Performance of C++ implementation and deconvlucy wascompared

� Super-resolution algorithm was implemented in C++

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm

Page 37: On Comparative study of Image Restoration Algorithms

References

W. H. Richardson, �Bayesian-based iterative method of imagerestoration,� Journal of the Optical Society of America, vol. 62,pp. 55�59, January 1972.

L. B. Lucy, �An iteration technique for the recti�cation of theobscured distribution,� The Astronomical Journal, vol. 79,pp. 745�754, June 1974.

D. S. C. Biggs and M. Andrews, �Acceleration of iterativeimage restoration algorithms,� Applied Optics, vol. 36, no. 8,pp. 1766�1775, 1997.

H. Bi and G. Boerner, �When does the Richardson-Lucydeconvolution converge?,� Astronomy and Astrophysics Suppl.,vol. 108, pp. 409�415, Dec. 1994.

Faizan Abbas Ali Comparative Study of Image Restoration Algorithm