Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum.

Slides:



Advertisements
Similar presentations
Inferring the kernel: multiscale method Input image Loop over scales Variational Bayes Upsample estimates Use multi-scale approach to avoid local minima:
Advertisements

Removing blur due to camera shake from images. William T. Freeman Joint work with Rob Fergus, Anat Levin, Yair Weiss, Fredo Durand, Aaron Hertzman, Sam.
Bayesian Belief Propagation
S INGLE -I MAGE R EFOCUSING AND D EFOCUSING Wei Zhang, Nember, IEEE, and Wai-Kuen Cham, Senior Member, IEEE.
NASSP Masters 5003F - Computational Astronomy Lecture 5: source detection. Test the null hypothesis (NH). –The NH says: let’s suppose there is no.
Patch-based Image Deconvolution via Joint Modeling of Sparse Priors Chao Jia and Brian L. Evans The University of Texas at Austin 12 Sep
CSC321: 2011 Introduction to Neural Networks and Machine Learning Lecture 10: The Bayesian way to fit models Geoffrey Hinton.
Reducing Drift in Parametric Motion Tracking
Shaojie Zhuo, Dong Guo, Terence Sim School of Computing, National University of Singapore CVPR2010 Reporter: 周 澄 (A.J.) 01/16/2011 Key words: image deblur,
Unnatural L 0 Representation for Natural Image Deblurring Speaker: Wei-Sheng Lai Date: 2013/04/26.
Removing Camera Shake from a Single Photograph 报告人:牟加俊 日期: In ACM SIGGRAPH, 2006.
1 Removing Camera Shake from a Single Photograph Rob Fergus, Barun Singh, Aaron Hertzmann, Sam T. Roweis and William T. Freeman ACM SIGGRAPH 2006, Boston,
Rob Fergus Courant Institute of Mathematical Sciences New York University A Variational Approach to Blind Image Deconvolution.
Optimization & Learning for Registration of Moving Dynamic Textures Junzhou Huang 1, Xiaolei Huang 2, Dimitris Metaxas 1 Rutgers University 1, Lehigh University.
Personal Photo Enhancement using Example Images Neel Joshi Wojciech Matusik, Edward H. Adelson, and David J. Kriegman Microsoft Research, Disney Research,
ECE 472/572 - Digital Image Processing Lecture 8 - Image Restoration – Linear, Position-Invariant Degradations 10/10/11.
EE465: Introduction to Digital Image Processing
Computer Vision - A Modern Approach
Machine Learning CUNY Graduate Center Lecture 3: Linear Regression.
Interactive Matting Christoph Rhemann Supervised by: Margrit Gelautz and Carsten Rother.
Image Deblurring with Optimizations Qi Shan Leo Jiaya Jia Aseem Agarwala University of Washington The Chinese University of Hong Kong Adobe Systems, Inc.
Our output Blur kernel. Close-up of child Our output Original photograph.
Miguel Lourenço, João P. Barreto, Abed Malti Institute for Systems and Robotics, Faculty of Science and Technology University of Coimbra, Portugal Feature.
Understanding and evaluating blind deconvolution algorithms
(1) A probability model respecting those covariance observations: Gaussian Maximum entropy probability distribution for a given covariance observation.
Rician Noise Removal in Diffusion Tensor MRI
Automatic Estimation and Removal of Noise from a Single Image
Noise Estimation from a Single Image Ce Liu William T. FreemanRichard Szeliski Sing Bing Kang.
ELE 488 F06 ELE 488 Fall 2006 Image Processing and Transmission ( ) Wiener Filtering Derivation Comments Re-sampling and Re-sizing 1D  2D 10/5/06.
High dynamic range imaging. Camera pipeline 12 bits8 bits.
1 Bayesian methods for parameter estimation and data assimilation with crop models Part 2: Likelihood function and prior distribution David Makowski and.
1 Patch Complexity, Finite Pixel Correlations and Optimal Denoising Anat Levin, Boaz Nadler, Fredo Durand and Bill Freeman Weizmann Institute, MIT CSAIL.
Machine Learning CUNY Graduate Center Lecture 3: Linear Regression.
© by Yu Hen Hu 1 ECE533 Digital Image Processing Image Restoration.
Computer Vision - Restoration Hanyang University Jong-Il Park.
Chapter 7 Case Study 1: Image Deconvolution. Different Types of Image Blur Defocus blur --- Depth of field effects Scene motion --- Objects in the scene.
Deconvolution, Deblurring and Restoration T , Biomedical Image Analysis Seminar Presentation Seppo Mattila & Mika Pollari.
ECE 8443 – Pattern Recognition ECE 8423 – Adaptive Signal Processing Objectives: Deterministic vs. Random Maximum A Posteriori Maximum Likelihood Minimum.
Yu-Wing Tai, Hao Du, Michael S. Brown, Stephen Lin CVPR’08 (Longer Version in Revision at IEEE Trans PAMI) Google Search: Video Deblurring Spatially Varying.
ECE 8443 – Pattern Recognition LECTURE 07: MAXIMUM LIKELIHOOD AND BAYESIAN ESTIMATION Objectives: Class-Conditional Density The Multivariate Case General.
Image Processing Edge detection Filtering: Noise suppresion.
EDGE DETECTION IN COMPUTER VISION SYSTEMS PRESENTATION BY : ATUL CHOPRA JUNE EE-6358 COMPUTER VISION UNIVERSITY OF TEXAS AT ARLINGTON.
Image deconvolution, denoising and compression T.E. Gureyev and Ya.I.Nesterets
Learning to Perceive Transparency from the Statistics of Natural Scenes Anat Levin School of Computer Science and Engineering The Hebrew University of.
8-1 Chapter 8: Image Restoration Image enhancement: Overlook degradation processes, deal with images intuitively Image restoration: Known degradation processes;
Vincent DeVito Computer Systems Lab The goal of my project is to take an image input, artificially blur it using a known blur kernel, then.
ECE 8443 – Pattern Recognition ECE 8527 – Introduction to Machine Learning and Pattern Recognition LECTURE 07: BAYESIAN ESTIMATION (Cont.) Objectives:
INTRODUCTION TO Machine Learning 3rd Edition
1 Markov random field: A brief introduction (2) Tzu-Cheng Jen Institute of Electronics, NCTU
- 1 - Overall procedure of validation Calibration Validation Figure 12.4 Validation, calibration, and prediction (Oberkampf and Barone, 2004 ). Model accuracy.
Lecture#4 Image reconstruction
Lecture 8 Source detection NASSP Masters 5003S - Computational Astronomy
Machine Learning 5. Parametric Methods.
Chapter 5 Image Restoration.
Removing motion blur from a single image
Univariate Gaussian Case (Cont.)
Vincent DeVito Computer Systems Lab The goal of my project is to take an image input, artificially blur it using a known blur kernel, then.
Machine Vision Edge Detection Techniques ENT 273 Lecture 6 Hema C.R.
Today Defocus Deconvolution / inverse filters. Defocus.
ICCV 2007 Optimization & Learning for Registration of Moving Dynamic Textures Junzhou Huang 1, Xiaolei Huang 2, Dimitris Metaxas 1 Rutgers University 1,
RECONSTRUCTION OF MULTI- SPECTRAL IMAGES USING MAP Gaurav.
Ch 1. Introduction Pattern Recognition and Machine Learning, C. M. Bishop, Updated by J.-H. Eom (2 nd round revision) Summarized by K.-I.
Degradation/Restoration Model
Image Deblurring and noise reduction in python
A Neural Approach to Blind Motion Deblurring
Deconvolution , , Computational Photography
Removing motion blur from a single image
Advanced deconvolution techniques and medical radiography
Deblurring Shaken and Partially Saturated Images
Presentation transcript:

Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum

Agenda 1.Problem Statement 2.Introduction to Non-Blind Deconvolution 3.Solutions & Approaches A.Image Deblurring PSF Estimation using Sharp Edge Prediction / Neel Joshi et. Al. B.MAP x,k Solution Analysis Understanding and evaluating blind deconvolution algorithms / Anat Levin et. Al. C. Variational Method MAP k Removing Camera Shake from a Single Photograph / Rob Fergus et. Al 4.Summary 2

Problem statement Blur = Degradation of sharpness and contrast of the image, causing loss of high frequencies. Technically - convolution with certain kernel during the imaging process. 3

Camera Motion blur 4

Defocus blur 5

6

7

Blur – generative model = = Point Spread Function Optical Transfer Function fft(Image) Sharp imageBlured image fft(Blured image)

Object Motion blur 9

Local Camera Motion 10

Depth of field – Local defocus 11

Lucy Richardson Evolution of algorithms ? Camera motion blur Simple kernels Non blind deconvolution Wiener 1949 Joshi 2008 Shan 2008 Volunteers ? Fergus 2006

Introduction to Non-Blind Deconvolution blur kernelblurred image sharp image Deconvolution Evolution: Simple no-Noise Case Noise Effect Over Simple Solution Wiener Deconvolution RL Deconvolution noise 13

Simple no-Noise Case: BlurredImageRecovered 14

Noisy case: 15 Original (x)Blured + noise (y)Recovered x

Original signal FT of original signal Convolved signals w/o noise FT of convolved signals sd Reconstructed FT of the signal High Frequency Noise Amplified 16 Noisy case, 1D Example: Noisy Signal Original Signal

Regularizing by window Window size:

Wiener Deconvolution Blurred noisy image Recovered image 18

Non Blind Iterative Method : Richardson –Lucy Algorithm Assumptions: Blurred image y i ~P(y i ), Sharp image x j ~P(x j ) i point in y, j point in x Target: Recover P(x) given P(y) & P(y|x) From Bayes theorem Object distribution can be expressed iteratively: Richardson, W.H., “Bayesian-Based Iterative Method of Image Restoration”, J. Opt. Soc. Am., 62, 55, (1972). Lucy, L.B., “An iterative technique for the rectification of observed distributions”, Astron. J., 79, 745, (1974). 19 where

Richardson-Lucy Application Simulated Multiple Star measurement PSF Identification reconstruction of 4 th Element 20

21

Richardson-Lucy iterative approach (examples) Blurred noisy image 10 iterations 50 iterations 100 iterations 22

Solution Approaches A.Image Deblurring PSF Estimation using Sharp Edge Prediction Neel Joshi Richard Szeliski David J. Kriegman B.MAP x,k Solution Analysis Understanding and evaluating blind deconvolution algorithms Anat Levin, Yair Weiss, Fredo Durand, William T. Freeman C. Variational Method MAP k Removing Camera Shake from a Single Photograph Rob Fergus, Barun Singh, Aaron Hertzmann, Sam T. Roweis, William T. Freeman 23

PSF Estimation by Sharp Edge Prediction Given edge steps, debluring can be reduced to Kernel Optimization Suggested in PSF Estimation by Sharp Edge Prediction \ Neel Joshi et. el. in Select Edge Step (Masking) Estimate Blurring Kernel Recover Latent Image 24

PSF Estimation by Sharp Edge Prediction - Masking Original ImageEdge Prediction Masking Min Max Valid Region 25

Masking, Cont. Which is Best the Signals ? Edge Impulse Original Blurred 26

Blurr Model: y=x*k+n, n ~ N( 0,σ 2 ) Bayseian Framework: P(k|y) = P(y|k)P(k)/P(y) Map Model: argmax k P(k|y) = argmin k L(y|k) + L(k) PSF Estimation by Sharp Edge Prediction – PSF Estimation 27

PSF Estimation by Sharp Edge Prediction – Recovery Recovery through Lucy-Richardson Iterations given the PSF kernel 28 BlurredRecovered

PSF Estimation by Sharp Edge, Summary & Improvements 1.Handle RGB Images – perform processing in parallel 2.Local Kernel Variations: Sub divide image into sub-image units Limitations: – Highly depends on the quality of the edge detection – Requires Strong Edges in multiple orientations for proper kernel estimation – Assumes knowledge of noise error figure. 29

blur kernel MAP x,k, Blind Deconvolution Definition blurred image sharp image noise Input (known) Unknown, need to estimate ? ? Courtesy of Anat Levin CVPR 09 Slides30

MAP x,k Cont. - Natural Image Priors Derivative histogram from a natural image Parametric models Derivative distributions in natural images are sparse: Log prob x x Gaussian: -x 2 Laplacian: - |x| -|x| 0.5 -|x| 0.25 Courtesy of Anat Levin CVPR 09 Slides31

Naïve MAP x,k estimation Given blurred image y, Find a kernel k and latent image x minimizing: Should favor sharper x explanations Convolution constraint Sparse prior Courtesy of Anat Levin CVPR 09 Slides32

The MAP x,k paradox P (, ) >P (, ) Let be an arbitrarily large image sampled from a sparse prior, and Then the delta explanation is favored Latent image kernel Latent image kernel Courtesy of Anat Levin CVPR 09 Slides33

? The MAP x,k failure sharpblurred Courtesy of Anat Levin CVPR 09 Slides34

The MAP x,k failure Red windows = [ p(sharp x) >p(blurred x) ] 15x15 windows 25x25 windows 45x45 windows simple derivatives [-1,1],[-1;1] FoE filters (Roth&Black) 35

P (blurred step edge) sum of derivatives: cheaper The MAP x,k failure - intuition P (blurred impulse) P (impulse) sum of derivatives: cheaper > P (step edge) < k=[0.5,0.5] Courtesy of Anat Levin CVPR 09 Slides36

P (blurred real image) MAP x,k Cont. - Blur Reduces Derivative Contrast Noise and texture behave as impulses - total derivative contrast reduced by blur < P (sharp real image) cheaper Courtesy of Anat Levin CVPR 09 Slides37

MAP x,k Reweighting Solution Alternating Optimization Between x & k Minimization term: MAP x,k High Quality Motion Debluring From Single Image / Shan et al.

39 MAP x,k Reweighting - Blurred 39

40 MAP x,k Reweighting - Recovered

MAP x,k Reweighting – Cont. 41 Coarse –to-Fine Recovery

MAP x,k Reweighting Intuition 42 w i high value penalty for smoooth areas → Momentum for escaping delta kernel λ Increases through iterations → “Cooling Effect” increasing Probability for Halting 0=λ Sharper image on the horizon ∞=λ Further iterations not possible

44

45 Example 2 45

46 Output 2 46

Solution Approaches A.Image Deblurring PSF Estimation using Sharp Edge Prediction Neel Joshi Richard Szeliski David J. Kriegman B.MAP x,k Solution Analysis Understanding and evaluating blind deconvolution algorithms Anat Levin, Yair Weiss, Fredo Durand, William T. Freeman C. Variational Method MAP k Removing Camera Shake from a Single Photograph Rob Fergus, Barun Singh, Aaron Hertzmann, Sam T. Roweis, William T. Freeman 47

MAP k estimation Given blurred image y, Find a kernel minimizing: Again, Should favor sharper x explanations Convolution constraint Sparse priorKernel prior 48

Superiority of MAP k over MAP k,x Toy Problem : y=kx+n The joint distribution p(x, k|y). Maximum for x → 0, k → ∞. p(k|y) produce optimum closer to true k ∗. uncertainty of p(k|y) reduces given multiple observations y j =kx j + n j. 49

Evaluation on 1D signals MAP k variational approximation (Fergus et al.) Exact MAP k MAP x,k Favors delta solution MAP k Gaussian prior Favor correct solution despite wrong prior! Courtesy of Anat Levin CVPR 09 Slides 50

Intuition: dimensionality asymmetry MAP x,k – Estimation unreliable. Number of measurements always lower than number of unknowns: #y<#x+#k MAP k – Estimation reliable. Many measurements for large images: #y>>#k Large, ~10 5 unknownsSmall, ~10 2 unknowns blurred image y kernel k sharp image x ~10 5 measurements Courtesy of Anat Levin CVPR 09 Slides51

Courtesy of Rob Fergus Slides52

Three sources of information Courtesy of Rob Fergus Slides53

Likelihood p(y|b,x) Courtesy of Rob Fergus Slides54

Image prior p(x) Courtesy of Rob Fergus Slides55

Blur prior p(b) Courtesy of Rob Fergus Slides56

The obvious thing to do Courtesy of Rob Fergus Slides57

Variational Bayesian approach Courtesy of Rob Fergus Slides58

Variational Bayesian methods Variational Bayesian = ensemble learning, A family of techniques for approximating intractable integrals arising in Bayesian inference and machine learning. Lower bound the marginal likelihood (i.e. "evidence") of several models with a view to performing model selection. 59

Setup of Variational Approach

Ensemble Learning for Blind Source Separation / J.W. Miskin, D.J.C. MacKay Small synthetic blurs large real world blurs Cartoon images Gradients of natural images Independent Factor Analysis \ H. Attias An introduction to variational methods for graphical models \ JORDAN M. et al. 61

63

64

65

Courtesy of Rob Fergus Slides66

Example 1 67

Output 1 68

Example 2 69

Output 2 70

Example 3 71

Output 3 72

Achievements Work on real world images Deals with large camera motions (up to 60 pixels) Getting close to practical generic solution of an old problem. 73

Limitations Targeted at camera motion blur – No in plane rotation – No motion in picture – Out of focus blur Manual input – Region of Interest – Kernel size & orientation – Other parameters e.g. scale offset, kernel TH & 9 other semi-fixed parameters Sensitive to image compression, noise(dark images) & saturation Still contains artifacts (solvable by upgrading from Lucy Richardson) 74

Evaluation Cumulative histogram of deconvolution successes : bin r = #{ deconv error < r } MAP k, Gaussian prior Shan et al. SIGGRAPH08 Fergus, variational MAP k MAP x,k sparse prior Successes percent 75

Ground truth data acquisition 4 images x 8 kernels = 32 test images Data available online: 76

Fergus et al. SIGGRAPH06 MAP k, variational approx. Comparison Shan et al. SIGGRAPH08 adjusted MAP x,k MAP x,k MAP k, Gaussian prior Ground truth 77

Summary Variational MAP k Fergus Reweighted MAP KX Shan Quasi-MAP K Joshi Method Camera motion blur Complex sparse PSF Defocus blur simple PSF Distortion model User selectedEdge region Region of interest Variational Bayes for K estimation (MAP k equivalent) MAP KX Quasi-MAP K Optimization model O(K+X prior +  PRIOR ) O(K+X)O(K)Degrees of freedom Multiscale iterative (internal altering) Alternating iterativeGradient based least squares Scheme 78

Conclusion reduce # of estimated parameters using – Priors. – Kernel marginalization Seperating the problem into kernel recovery & non-blind deconvolution Existing challenges & potential research – Solutions to spatially varying kernels – Robustness to user’s parameters & initial priors Debluring single image under constrained problem

Debluring is underconstrained Debluring single image under constrained problem ? Blured image Recovered image Recovered kernel

Priors do the trick ? Blured image Image prior Recovered kernel

Kernel marginalization ? Blured image Recovered kernel Image prior

Back to non-blind deconvolution ? Recovered imageBlured image Recovered kernel

Existing challenges and potential research Robustness to user’s parameters & initial priors Solutions to spatially varying kernels 84

Thank You Eitan & Tomer The End

87