single image blind deconvolution presented by: tomer peled & eitan shterenbaum
TRANSCRIPT
![Page 1: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/1.jpg)
Single Image Blind Deconvolution
Presented By:Tomer Peled
&Eitan Shterenbaum
![Page 2: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/2.jpg)
Agenda
1. Problem Statement2. Introduction to Non-Blind Deconvolution3. Solutions & Approaches
A. Image Deblurring PSF Estimation using Sharp Edge Prediction / Neel Joshi et. Al.
B. MAPx,k Solution AnalysisUnderstanding and evaluating blind deconvolution algorithms / Anat Levin et. Al.
C. Variational Method MAPkRemoving Camera Shake from a Single Photograph / Rob Fergus et. Al
4. Summary2
![Page 3: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/3.jpg)
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
![Page 4: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/4.jpg)
Camera Motion blur
4
![Page 5: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/5.jpg)
Defocus blur
5
![Page 6: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/6.jpg)
Defocus blur
6
![Page 7: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/7.jpg)
Defocus blur
7
![Page 8: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/8.jpg)
Blur – generative model
=
=
Point Spread Function
Optical Transfer Functionfft(Image)
Sharp image Blured image
fft(Blured image)
![Page 9: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/9.jpg)
Object Motion blur
9
![Page 10: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/10.jpg)
Local Camera Motion
10
![Page 11: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/11.jpg)
Depth of field – Local defocus
11
![Page 12: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/12.jpg)
Lucy Richardson
Evolution of algorithms
?
Camera motion blur
Simple kernels
Non blind deconvolution
12
1972
Wiener1949
Joshi2008
Shan2008
Volunteers ?
Fergus2006
![Page 13: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/13.jpg)
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
![Page 14: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/14.jpg)
Simple no-Noise Case:
BlurreBlurredd
ImageImage
RecoveredRecovered
14
![Page 15: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/15.jpg)
Noisy case:
15
Original (x) Blured + noise (y) Recovered x
![Page 16: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/16.jpg)
Original signalOriginal signalFT of original signalFT of original signal
Convolved signals w/o noiseConvolved signals w/o noiseFT of convolved signalsFT of convolved signals sd
Reconstructed FT of the Reconstructed FT of the signalsignal
High Frequency Noise Amplified16
Noisy case, 1D Example:
Noisy SignalOriginal Signal
![Page 17: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/17.jpg)
Wiener Deconvolution
Blurred noisy Blurred noisy imageimage
Recovered imageRecovered image
18
![Page 18: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/18.jpg)
Non Blind Iterative Method : Richardson –Lucy Algorithm
Assumptions: Blurred image yi~P(yi), Sharp image xj~P(xj) 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
![Page 19: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/19.jpg)
Richardson-Lucy ApplicationSimulated Multiple Star
measurement PSF Identification reconstruction of 4th Element
20
![Page 20: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/20.jpg)
Solution Approaches
A. Image Deblurring PSF Estimation using Sharp Edge PredictionNeel Joshi Richard Szeliski David J. Kriegman
B. MAPx,k Solution AnalysisUnderstanding and evaluating blind deconvolution algorithmsAnat Levin, Yair Weiss, Fredo Durand, William T. Freeman
C. Variational Method MAPkRemoving Camera Shake from a Single PhotographRob Fergus, Barun Singh, Aaron Hertzmann, Sam T. Roweis, William T. Freeman
23
![Page 21: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/21.jpg)
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
![Page 22: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/22.jpg)
PSF Estimation by Sharp Edge Prediction - Masking
Original Image Edge Prediction Masking
Min
Max
Valid Region
25
![Page 23: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/23.jpg)
Masking, Cont.Which is Best the Signals?
Edge
Impulse
Original Blurred
26
![Page 24: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/24.jpg)
Blurr Model: y=x*k+n, n ~ N(0,σ2)
Bayseian Framework: P(k|y) = P(y|k)P(k)/P(y)
Map Model:argmaxk P(k|y) = argmink L(y|k) + L(k)
PSF Estimation by Sharp Edge Prediction – PSF Estimation
27
![Page 25: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/25.jpg)
PSF Estimation by Sharp Edge Prediction – Recovery
Recovery through Lucy-Richardson Iterations given the PSF kernel
28
Blurred Recovered
![Page 26: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/26.jpg)
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
![Page 27: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/27.jpg)
blur kernel
MAPx,k , Blind Deconvolution Definition
blurred image sharp image
noise
Input (known)
Unknown, need to estimate
?
?Courtesy of Anat Levin CVPR 09 Slides30
![Page 28: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/28.jpg)
MAPx,k Cont. - Natural Image Priors
Derivative histogram from a natural image
Parametric models
Derivative distributions in natural images are sparse:
Log
prob
xx
Gaussian:
-x2
Laplacian:
-|x||-x|0.5
|-x|0.25
Courtesy of Anat Levin CVPR 09 Slides31
![Page 29: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/29.jpg)
Naïve MAPx,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
![Page 30: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/30.jpg)
The MAPx,k paradox
P( , )>P ),( Let be an arbitrarily large image sampled from a sparse prior , and
Then the delta explanation is favored
Latent imagekernel
Latent imagekernel
Courtesy of Anat Levin CVPR 09 Slides33
![Page 31: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/31.jpg)
?
The MAPx,k failure sharp blurred
Courtesy of Anat Levin CVPR 09 Slides34
![Page 32: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/32.jpg)
The MAPx,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
![Page 33: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/33.jpg)
P(blurred step edge)
sum of derivatives: cheaper
The MAPx,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
![Page 34: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/34.jpg)
P(blurred real image)
MAPx,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
![Page 35: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/35.jpg)
MAPx,k Reweighting Solution
Alternating Optimization Between x & k
Minimization term:
MAPx,k
High Quality Motion Debluring From Single Image / Shan et al.
![Page 36: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/36.jpg)
39
MAPx,k Reweighting - Blurred
39
![Page 37: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/37.jpg)
4040
MAPx,k Reweighting - Recovered
![Page 38: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/38.jpg)
Solution Approaches
A. Image Deblurring PSF Estimation using Sharp Edge PredictionNeel Joshi Richard Szeliski David J. Kriegman
B. MAPx,k Solution AnalysisUnderstanding and evaluating blind deconvolution algorithmsAnat Levin, Yair Weiss, Fredo Durand, William T. Freeman
C. Variational Method MAPkRemoving Camera Shake from a Single PhotographRob Fergus, Barun Singh, Aaron Hertzmann, Sam T. Roweis, William T. Freeman
47
![Page 39: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/39.jpg)
MAPk estimation
Given blurred image y, Find a kernel minimizing:
Again, Should favor sharper x explanations
Convolution constraint
Sparse prior Kernel prior
48
![Page 40: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/40.jpg)
Superiority of MAPk over MAPk,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 yj =kxj + nj .
49
![Page 41: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/41.jpg)
Evaluation on 1D signals
MAPk variational approximation (Fergus et al.)
Exact MAPk MAPx,kFavors delta solution
MAPk Gaussian prior
Favor correct solution despite
wrong prior!
Courtesy of Anat Levin CVPR 09 Slides50
![Page 42: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/42.jpg)
Intuition: dimensionality asymmetry
MAPx,k– Estimation unreliable. Number of measurements always lower than number of unknowns: #y<#x+#k
MAPk – Estimation reliable. Many measurements for large images: #y>>#k
Large, ~105 unknowns Small, ~102 unknowns
blurred image ykernel k
sharp image x
~105 measurements
Courtesy of Anat Levin CVPR 09 Slides51
![Page 43: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/43.jpg)
Courtesy of Rob Fergus Slides52
![Page 44: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/44.jpg)
Three sources of information
Courtesy of Rob Fergus Slides53
![Page 45: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/45.jpg)
Image prior p(x)
Courtesy of Rob Fergus Slides55
![Page 46: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/46.jpg)
Blur prior p(b)
Courtesy of Rob Fergus Slides56
![Page 47: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/47.jpg)
The obvious thing to do
Courtesy of Rob Fergus Slides57
![Page 48: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/48.jpg)
Variational Bayesian approach
Courtesy of Rob Fergus Slides58
![Page 49: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/49.jpg)
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
![Page 50: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/50.jpg)
Setup of Variational Approach
![Page 51: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/51.jpg)
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. AttiasAn introduction to variational methods for graphical models \ JORDAN M. et al.
61
![Page 52: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/52.jpg)
![Page 53: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/53.jpg)
63
![Page 54: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/54.jpg)
64
![Page 55: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/55.jpg)
65
![Page 56: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/56.jpg)
Courtesy of Rob Fergus Slides66
![Page 57: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/57.jpg)
Example 1
67
![Page 58: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/58.jpg)
Output 1
68
![Page 59: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/59.jpg)
Example 2
69
![Page 60: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/60.jpg)
Output 2
70
![Page 61: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/61.jpg)
Example 3
71
![Page 62: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/62.jpg)
Output 3
72
![Page 63: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/63.jpg)
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
![Page 64: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/64.jpg)
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
![Page 65: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/65.jpg)
Evaluation
Cumulative histogram of deconvolution successes:
bin r = #{ deconv error > r }
MAPk, Gaussian prior
Shan et al. SIGGRAPH08Fergus, variational MAPk
MAPx,k sparse prior
100
80
60
40
20
Su
cces
ses
per
cen
t
75
![Page 66: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/66.jpg)
Summary
MethodQuasi-MAPK
JoshiReweighted MAPKX
ShanVariational MAPk
Fergus
Distortion modelDefocus blursimple PSF
Camera motion blurComplex sparse PSF
Camera motion blurComplex sparse PSF
Region of interestEdge regionEdge regionUser selected
Optimization modelQuasi-MAPKMAPKXVariational Bayes for K estimation (MAPk equivalent)
Degrees of freedomO(K)O(K+X)O(K+Xprior+PRIOR)
SchemeGradient based least squares
Alternating iterativeMultiscale iterative(internal altering)
78
![Page 67: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/67.jpg)
Debluring is underconstrained
Debluring single image under constrained
problem
?Blured imageRecovered image
Recovered kernel
![Page 68: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/68.jpg)
Priors do the trick
?Blured image
Image prior
Recovered kernel
![Page 69: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/69.jpg)
Kernel marginalization
?Blured image
Recovered kernel
Image prior
![Page 70: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/70.jpg)
Back to non-blind deconvolution
?Recovered image Blured image
Recovered kernel
![Page 71: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/71.jpg)
Existing challenges and potential research
• Robustness to user’s parameters & initial priors
• Solutions to spatially varying kernels
84
![Page 72: Single Image Blind Deconvolution Presented By: Tomer Peled & Eitan Shterenbaum](https://reader035.vdocuments.site/reader035/viewer/2022062804/56649d045503460f949d7dc3/html5/thumbnails/72.jpg)
Thank You Eitan & Tomer
The End