Optimal Denoising of Natural Images and their Multiscale Geometry and Density

Size: px
Start display at page:

Download "Optimal Denoising of Natural Images and their Multiscale Geometry and Density"

Transcription

1 Optimal Denoising of Natural Images and their Multiscale Geometry and Density Department of Computer Science and Applied Mathematics Weizmann Institute of Science, Israel. Joint work with Anat Levin (WIS), Fredo Durand and Bill Freeman (MIT). June 2013 [Levin and Nadler, CVPR 2011] [Levin, Nadler, Durand, Freeman, ECCV 2012]

2 Classical Statistical Estimation vs. Image Restoration Classical Statistical Inference: {x i } n i=1 are n i.i.d. random samples (observations) from some unknown density p(x) (in parametric setting, p(x) = p θ (x) ). Task: Given {x i } wish to estimate p(x) (or θ).

3 Classical Statistical Estimation vs. Image Restoration Classical Statistical Inference: {x i } n i=1 are n i.i.d. random samples (observations) from some unknown density p(x) (in parametric setting, p(x) = p θ (x) ). Task: Given {x i } wish to estimate p(x) (or θ). Key result: Consistency As n, ˆp(x) p(x), and ˆθ θ

4 Restoration Tasks and Image Priors Many image restoration tasks: denoising, super-resolution, deblurring, etc, are different:

5 Restoration Tasks and Image Priors Many image restoration tasks: denoising, super-resolution, deblurring, etc, are different: Have only one (or even less) noisy observation per each unknown.

6 Restoration Tasks and Image Priors Many image restoration tasks: denoising, super-resolution, deblurring, etc, are different: Have only one (or even less) noisy observation per each unknown. Image Restoration Tasks are ill-posed problems.

7 Restoration Tasks and Image Priors Many image restoration tasks: denoising, super-resolution, deblurring, etc, are different: Have only one (or even less) noisy observation per each unknown. Image Restoration Tasks are ill-posed problems. Given a degraded image y, there are multiple valid explanations x.

8 Restoration Tasks and Image Priors Many image restoration tasks: denoising, super-resolution, deblurring, etc, are different: Have only one (or even less) noisy observation per each unknown. Image Restoration Tasks are ill-posed problems. Given a degraded image y, there are multiple valid explanations x. Natural image priors are crucial to rule out unlikely solutions.

9 On the Strength of Natural Image Priors

10 On the Strength of Natural Image Priors Natural k k patches are extremely sparse in the set of 256 k k patches

11 Optimal Image Restoration Practical / Computational issues limit most current image priors to local patches.

12 Optimal Image Restoration Practical / Computational issues limit most current image priors to local patches. Hence, the restoration results are suboptimal.

13 Towards Optimal Image Restoration

14 Towards Optimal Image Restoration Fundamental Theoretical Questions:

15 Towards Optimal Image Restoration Fundamental Theoretical Questions: - How accurate are currently used priors?

16 Towards Optimal Image Restoration Fundamental Theoretical Questions: - How accurate are currently used priors? - How much gain is possible by increasing patch size?

17 Towards Optimal Image Restoration Fundamental Theoretical Questions: - How accurate are currently used priors? - How much gain is possible by increasing patch size? - Computational issues aside, what is the optimal possible restoration? Does it have a zero error?

18 Towards Optimal Image Restoration Fundamental Theoretical Questions: - How accurate are currently used priors? - How much gain is possible by increasing patch size? - Computational issues aside, what is the optimal possible restoration? Does it have a zero error? - How do these questions relate to natural image statistics?

19 Towards Optimal Image Restoration Fundamental Theoretical Questions: - How accurate are currently used priors? - How much gain is possible by increasing patch size? - Computational issues aside, what is the optimal possible restoration? Does it have a zero error? - How do these questions relate to natural image statistics? This talk will try to address some of these issues for a specific task: denoising

20 Outline Denoising of Natural Images: - Ill posed problem - solved using priors - Optimal prior = (unknown) density of natural images

21 Outline Denoising of Natural Images: - Ill posed problem - solved using priors - Optimal prior = (unknown) density of natural images Questions:

22 Outline Denoising of Natural Images: - Ill posed problem - solved using priors - Optimal prior = (unknown) density of natural images Questions: i) What is the optimal achievable denoising? How is it related to the multiscale geometry and density of natural images?

23 Outline Denoising of Natural Images: - Ill posed problem - solved using priors - Optimal prior = (unknown) density of natural images Questions: i) What is the optimal achievable denoising? How is it related to the multiscale geometry and density of natural images? ii) How far from optimal are current algorithms with approximate priors?

24 Outline Denoising of Natural Images: - Ill posed problem - solved using priors - Optimal prior = (unknown) density of natural images Questions: i) What is the optimal achievable denoising? How is it related to the multiscale geometry and density of natural images? ii) How far from optimal are current algorithms with approximate priors? iii) How can we further improve current algorithms?

25 Image Denoising - The Final Frontier [Is denoising dead? / Chatterjee & Milanfar]

26 Image Denoising - The Final Frontier [Is denoising dead? / Chatterjee & Milanfar] Should you do your Ph.D. in image denoising?

27 Image Denoising - The Final Frontier [Is denoising dead? / Chatterjee & Milanfar] Should you do your Ph.D. in image denoising? Answer at end of talk...

28 Natural Image Denoising Problem Setting: x : unknown noise free natural image y = x + n : observed image corrupted by noise. In this talk, n - additive iid Gaussian N(0, σ 2 ).

29 Natural Image Denoising Problem Setting: x : unknown noise free natural image y = x + n : observed image corrupted by noise. In this talk, n - additive iid Gaussian N(0, σ 2 ). Task: Given y, construct clean version ŷ (estimate of x).

30 Natural Image Denoising Problem Setting: x : unknown noise free natural image y = x + n : observed image corrupted by noise. In this talk, n - additive iid Gaussian N(0, σ 2 ). Task: Given y, construct clean version ŷ (estimate of x). Performance Measure: MSE = E[ x ŷ 2 ]

31 Natural Image Denoising Problem Setting: x : unknown noise free natural image y = x + n : observed image corrupted by noise. In this talk, n - additive iid Gaussian N(0, σ 2 ). Task: Given y, construct clean version ŷ (estimate of x). Performance Measure: MSE = E[ x ŷ 2 ] Challenge: Ill posed problem: #unknowns = #observations

32 Denoising Algorithms Priors: All denoising algorithms employ implicitly or explicitly some prior on the unknown (noise-free) image. Many different priors and denoising algorithms s - Gabor filters, anisotropic diffusion 1990 s - wavelet based methods, total variation 2000 s - sparse representations,

33 Denoising Algorithms Priors: All denoising algorithms employ implicitly or explicitly some prior on the unknown (noise-free) image. Many different priors and denoising algorithms s - Gabor filters, anisotropic diffusion 1990 s - wavelet based methods, total variation 2000 s - sparse representations, 05 Non-Local Means Algorithm

34 The NL-Means Algorithm L. Yaroslavsky already suggested similar ideas in [Buades & al 2005] [Awate & Whittaker 2006] [Barash & Comaniciu 2004] where ŷ(x i ) = 1 D(x i ) D(x i ) = pixels j pixels j K ε (y(x i ), y(x j ))y(x j ) K ε (y(x i ), y(x j ))

35 The NL-Means Algorithm L. Yaroslavsky already suggested similar ideas in [Buades & al 2005] [Awate & Whittaker 2006] [Barash & Comaniciu 2004] where In matrix form ŷ(x i ) = 1 D(x i ) D(x i ) = pixels j pixels j K ε (y(x i ), y(x j ))y(x j ) K ε (y(x i ), y(x j )) ŷ = D 1 Wy One interpretation: Random walk on image patches. [Singer, Shkolnisky, N. 08 ]

36 Example: Input Image

37 Example: Original noise-free image

38 Example: Result of BM3D

39 Example: Result of KSVD

40 Image Denoising, Smoothness, Sparsity, Priors Key Questions About Natural Image Denoising:

41 Image Denoising, Smoothness, Sparsity, Priors Key Questions About Natural Image Denoising: - Optimal Denoising in this setup? Relation to geometry and density of natural image patches?

42 Image Denoising, Smoothness, Sparsity, Priors Key Questions About Natural Image Denoising: - Optimal Denoising in this setup? Relation to geometry and density of natural image patches? - Are we there yet? How much better can we improve on BM3D or other denoising algorithms?

43 Statistical Framework First consider denoising algorithms that estimate the central pixel x c from a k k window around it, dimension d = k 2. In other words, think of x, y as both k k patches. Goal: Given y = x + n, estimate central pixel of x as best as possible. Quality Measure: Mean Squared Error (MSE) MSE = E[(ŷ c x c ) 2 ], PSNR = 10 log 10 MSE

44 Optimal Natural Image Denoising The Bayesian Minimum Mean Squared Error Estimator µ(y) = E[x c y] = p(x c y)x c dx = p(y x)xc dx pd (x)p(y x)dx where p d (x) - density of d = k k patches of natural images.

45 Optimal Natural Image Denoising The Bayesian Minimum Mean Squared Error Estimator µ(y) = E[x c y] = p(x c y)x c dx = p(y x)xc dx pd (x)p(y x)dx where p d (x) - density of d = k k patches of natural images. NL-means ŷ(x i ) = j K(y(x j), y(x i ))y c (x j ) j K(y(x i), y(x j ))

46 Optimal Natural Image Denoising The Bayesian Minimum Mean Squared Error Estimator µ(y) = E[x c y] = p(x c y)x c dx = p(y x)xc dx pd (x)p(y x)dx where p d (x) - density of d = k k patches of natural images. NL-means ŷ(x i ) = j K(y(x j), y(x i ))y c (x j ) j K(y(x i), y(x j )) can be viewed as non-parametric approximation of the MMSE estimator using noisy image patches.

47 Upper and Lower Bounds on MMSE If p d (x) were known - in principle could compute MMSE and see how far we are from optimality (at fixed window size d).

48 Upper and Lower Bounds on MMSE If p d (x) were known - in principle could compute MMSE and see how far we are from optimality (at fixed window size d). Challenge: Derive lower bound on MMSE d w/out knowing p d (x)?

49 Upper and Lower Bounds on MMSE If p d (x) were known - in principle could compute MMSE and see how far we are from optimality (at fixed window size d). Challenge: Derive lower bound on MMSE d w/out knowing p d (x)? Solution: A trick...

50 A brute force / elegant non-parametric approach Don t know p d (x), but can sample from it.

51 A brute force / elegant non-parametric approach Don t know p d (x), but can sample from it. Approximate two identical yet different representations of MSE.

52 A brute force / elegant non-parametric approach Don t know p d (x), but can sample from it. Approximate two identical yet different representations of MSE. ERROR representation MSE = p(x) p(y x)(ˆµ(y) x c ) 2 dydx

53 A brute force / elegant non-parametric approach Don t know p d (x), but can sample from it. Approximate two identical yet different representations of MSE. ERROR representation MSE = p(x) p(y x)(ˆµ(y) x c ) 2 dydx VARIANCE representation MSE = p σ (y) p(x y)(ˆµ(y) x c ) 2 dxdy

54 A brute force / elegant non-parametric approach Don t know p d (x), but can sample from it.

55 A brute force / elegant non-parametric approach Don t know p d (x), but can sample from it. Consider a huge set of N = natural image patches. Approximate MMSE d non-parametrically. where for Gaussian noise and d = k 2. ˆµ(y) = p(y x) = 1 N 1 N i p(y x i)x i,c i p(y x, (1) i) 1 x y 2 (2πσ 2 e 2σ ) d/2 2 (2)

56 Upper and Lower Bound on MMSE Given a set of M noisy pairs {( x j, y j )} M j=1 and another independent set of N clean patches {x i } N i=1, both randomly sampled from natural images, we compute MMSE U = 1 (ˆµ(y j ) x j,c ) 2 (3) M j MMSE L = 1 ˆV(y j ) (4) M where ˆV(y j ) is the approximated variance: j ˆV(y j ) = 1 N i p(y j x i )(ˆµ(y j ) x i,c ) 2 1 N i p(y j x i ) (5)

57 Upper and Lower Bound on MMSE Given a set of M noisy pairs {( x j, y j )} M j=1 and another independent set of N clean patches {x i } N i=1, both randomly sampled from natural images, we compute MMSE U = 1 (ˆµ(y j ) x j,c ) 2 (3) M j MMSE L = 1 ˆV(y j ) (4) M where ˆV(y j ) is the approximated variance: j ˆV(y j ) = 1 N i p(y j x i )(ˆµ(y j ) x i,c ) 2 1 N i p(y j x i ) (5) Key Difference: MMSE U knows true noise-free patch x, MMSE L does not know x,

58 The Theoretical Part MMSE U is the MSE of yet another (not very fast) denoising algorithm. It is an upper bound on the optimal MSE.

59 The Theoretical Part MMSE U is the MSE of yet another (not very fast) denoising algorithm. It is an upper bound on the optimal MSE. Claim: In expectation, MMSE L is a lower bound on the MSE.

60 The Theoretical Part MMSE U is the MSE of yet another (not very fast) denoising algorithm. It is an upper bound on the optimal MSE. Claim: In expectation, MMSE L is a lower bound on the MSE. If N 1 is sufficiently large so that MMSE U MMSE L - then we have an accurate estimate of the MMSE!

61 The Theoretical Part Claim with ( ) 1 E N [ ˆV(y)] = V(y) + C(y)B(y) + o, (6) N B(y) = E[x 2 c y] 3E[x c y] 2 2E σ [x 2 c y] +4E[x c y]e σ [x c y] (7) C(y) = 1 p σ (y) N (4πσ 2 ) d/2 p(y) 2 (8) where p σ and E σ [ ] denote probability and expectation of random variables whose noise standard deviation is reduced from σ to σ = σ/ 2.

62 Theory Part II Next, we consider a local Gaussian approximation for p(x) around x = y, e.g., a Laplace approximation consisting of a second order Taylor expansion of ln p(x). For a Gaussian distribution analytic computation of bias Claim For a Gaussian distribution, B(y) 0 for all y.

63 Numerical Example 2 0 x 10 3 B(y j ) j

64 Experiment Design Set of N = natural image patches out of 20K images from LabelMe dataset. Independent set of M = 2000 patches from same dataset x. Add noise at different levels, σ = 18, 55, 170, and with different window sizes, k = 3, 4,..., 20. Compared MMSE L, MMSE U as well as MSE of various state-of-the-art denoising algorithms.

65 Experiments: σ = 18 PSNR MMSE U MMSE L LMMSE BM3D KSVD GSM Support size k

66 MMSE d vs window size PSNR patch size For small patches / large noise non-parametric approach can accurately estimate MMSE. [Levin and N. CVPR 2011]

67 MMSE d vs window size PSNR patch size Extrapolation: What happens as d?

68 Towards Results of Levin and Nadler [CVPR 2011] imply that with small patches current algorithms are close to optimal.

69 Towards Results of Levin and Nadler [CVPR 2011] imply that with small patches current algorithms are close to optimal. Open Questions:

70 Towards Results of Levin and Nadler [CVPR 2011] imply that with small patches current algorithms are close to optimal. Open Questions: -How much gain is possible by increasing patch size?

71 Towards Results of Levin and Nadler [CVPR 2011] imply that with small patches current algorithms are close to optimal. Open Questions: -How much gain is possible by increasing patch size? - Computational issues aside, what is the optimal possible restoration? Does it have a zero error?

72 Patch Complexity vs. PSNR gain Empirical Results: Assign each pixel to group G d where d is largest window size with reliable estimation.

73 Patch Complexity vs. PSNR gain Empirical Results: Assign each pixel to group G d where d is largest window size with reliable estimation. PSNR vs. number of pixels in window, for different groups G d 35 G 7 30 PSNR 30 G 11 G 14 G 17 G 21 G 25 G 30 PSNR 25 G 40 G Number of pixelsl Number of pixelsl 150 1D signals, σ = 10 2D signals, σ = G 49 G 69 G 113 G 149 G 197

74 Patch Complexity vs. PSNR gain Higher curves correspond to smooth regions, which flatten at larger patch dimensions, require smaller external database. Lower curves correspond to textured regions, which not only run out of samples sooner, but also their curves flatten earlier, and require huge external database.

75 Patch Complexity vs. PSNR gain Key conclusion: Law of Marginal Return

76 Patch Complexity vs. PSNR gain Key conclusion: Law of Marginal Return When increasing window size, Smooth regions: large gain, no need to increase sample size

77 Patch Complexity vs. PSNR gain Key conclusion: Law of Marginal Return When increasing window size, Smooth regions: large gain, no need to increase sample size Texture / edges: small gain, large increase in sample size

78 Adaptive Non-parametric Denoising For each pixel x i, find largest window w d for which ˆµ d is still reliable.

79 Adaptive Non-parametric Denoising For each pixel x i, find largest window w d for which ˆµ d is still reliable. σ Optimal Fixed Adaptive BM3D

80 Adaptive Non-parametric Denoising For each pixel x i, find largest window w d for which ˆµ d is still reliable. σ Optimal Fixed Adaptive BM3D Similar idea suggested for adaptive Non-Local Means and Boulanger. 2006]. [Kervrann Challenge: Develop Practical Algorithm Based on this Adaptivity Principle

81 PSNR vs. window size What is convergence rate of MMSE d to MMSE?

82 PSNR vs. window size What is convergence rate of MMSE d to MMSE? PSNR Data Polynomial fit Exponential fit d

83 PSNR vs. window size Empirically: MMSE d = e + c d α with α 1

84 PSNR vs. window size Empirically: MMSE d = e + c d α with α 1 Can this be predicted theoretically?

85 Dead Leaves Model Simple Image Formation Model: Dead Leaves [Matheron 68 ]

86 Dead Leaves Model Simple Image Formation Model: Dead Leaves [Matheron 68 ] Image = random collection of finite size piece-wise constant regions Region intensity = random variable with uniform distribution.

87 Multiscale Geometry of Natural Images Scale Invariance: many natural image statistics are scale invariant: Down-sampling natural images does not change gradient distribution, segment sizes, etc. [Ruderman 97, Field 78, etc.]

88 Multiscale Geometry of Natural Images Scale Invariance: many natural image statistics are scale invariant: Down-sampling natural images does not change gradient distribution, segment sizes, etc. [Ruderman 97, Field 78, etc.] Claim Under scale invariance Pr[pixel belongs to region of size s pixels ] 1/s [Alvarez, Gousseau, Morel, 99 ]

89 Scale Invariance and the dead leaves model Our Key Contribution: Claim Dead leaves model with scale invariance (and edge oracle) implies strictly positive MMSE and power law convergence with α = 1 MMSE d = e + c d α

90 Scale Invariance and the dead leaves model Our Key Contribution: Claim Dead leaves model with scale invariance (and edge oracle) implies strictly positive MMSE and power law convergence MMSE d = e + c d α with α = 1 Note that e = MMSE so we can extrapolate empirical curve and estimate PSNR.

91 Extrapolated PSNR vs. Current Algorithms σ Extrapolated bound KSVD BM3D EPLL

92 Extrapolated PSNR vs. Current Algorithms σ Extrapolated bound KSVD BM3D EPLL Still some modest room for improvement

93 Summary - Understanding the relation between natural image statistics and restoration taks is a fundamental problem.

94 Summary - Understanding the relation between natural image statistics and restoration taks is a fundamental problem. - Statistical framework to study optimal natural image denoising.

95 Summary - Understanding the relation between natural image statistics and restoration taks is a fundamental problem. - Statistical framework to study optimal natural image denoising. - Non-parametric methods - law of diminishing return.

96 Summary - Understanding the relation between natural image statistics and restoration taks is a fundamental problem. - Statistical framework to study optimal natural image denoising. - Non-parametric methods - law of diminishing return. - most further gain in smooth regions adaptive approach.

97 Summary - Understanding the relation between natural image statistics and restoration taks is a fundamental problem. - Statistical framework to study optimal natural image denoising. - Non-parametric methods - law of diminishing return. - most further gain in smooth regions adaptive approach. - Natural image statistics pose inherent limits on denoising. Empirically, power law convergence vs. window size, supported by Dead Leaves + Scale Invariance model.

98 Summary - Understanding the relation between natural image statistics and restoration taks is a fundamental problem. - Statistical framework to study optimal natural image denoising. - Non-parametric methods - law of diminishing return. - most further gain in smooth regions adaptive approach. - Natural image statistics pose inherent limits on denoising. Empirically, power law convergence vs. window size, supported by Dead Leaves + Scale Invariance model. - Still room for improvement (not by non-parametric method).

99 Summary - Understanding the relation between natural image statistics and restoration taks is a fundamental problem. - Statistical framework to study optimal natural image denoising. - Non-parametric methods - law of diminishing return. - most further gain in smooth regions adaptive approach. - Natural image statistics pose inherent limits on denoising. Empirically, power law convergence vs. window size, supported by Dead Leaves + Scale Invariance model. - Still room for improvement (not by non-parametric method). - Implications to other low level vision tasks - deconvolution, super-resolution.

100 Summary - Understanding the relation between natural image statistics and restoration taks is a fundamental problem. - Statistical framework to study optimal natural image denoising. - Non-parametric methods - law of diminishing return. - most further gain in smooth regions adaptive approach. - Natural image statistics pose inherent limits on denoising. Empirically, power law convergence vs. window size, supported by Dead Leaves + Scale Invariance model. - Still room for improvement (not by non-parametric method). - Implications to other low level vision tasks - deconvolution, super-resolution. THE END / THANK YOU!

Denoising an Image by Denoising its Components in a Moving Frame

Denoising an Image by Denoising its Components in a Moving Frame Denoising an Image by Denoising its Components in a Moving Frame Gabriela Ghimpețeanu 1, Thomas Batard 1, Marcelo Bertalmío 1, and Stacey Levine 2 1 Universitat Pompeu Fabra, Spain 2 Duquesne University,

More information

IMAGE RESTORATION VIA EFFICIENT GAUSSIAN MIXTURE MODEL LEARNING

IMAGE RESTORATION VIA EFFICIENT GAUSSIAN MIXTURE MODEL LEARNING IMAGE RESTORATION VIA EFFICIENT GAUSSIAN MIXTURE MODEL LEARNING Jianzhou Feng Li Song Xiaog Huo Xiaokang Yang Wenjun Zhang Shanghai Digital Media Processing Transmission Key Lab, Shanghai Jiaotong University

More information

Learning how to combine internal and external denoising methods

Learning how to combine internal and external denoising methods Learning how to combine internal and external denoising methods Harold Christopher Burger, Christian Schuler, and Stefan Harmeling Max Planck Institute for Intelligent Systems, Tübingen, Germany Abstract.

More information

IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING

IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING Idan Ram, Michael Elad and Israel Cohen Department of Electrical Engineering Department of Computer Science Technion - Israel Institute of Technology

More information

Sparse & Redundant Representations and Their Applications in Signal and Image Processing

Sparse & Redundant Representations and Their Applications in Signal and Image Processing Sparse & Redundant Representations and Their Applications in Signal and Image Processing Sparseland: An Estimation Point of View Michael Elad The Computer Science Department The Technion Israel Institute

More information

Markov Random Fields and Gibbs Sampling for Image Denoising

Markov Random Fields and Gibbs Sampling for Image Denoising Markov Random Fields and Gibbs Sampling for Image Denoising Chang Yue Electrical Engineering Stanford University changyue@stanfoed.edu Abstract This project applies Gibbs Sampling based on different Markov

More information

Locally Adaptive Regression Kernels with (many) Applications

Locally Adaptive Regression Kernels with (many) Applications Locally Adaptive Regression Kernels with (many) Applications Peyman Milanfar EE Department University of California, Santa Cruz Joint work with Hiro Takeda, Hae Jong Seo, Xiang Zhu Outline Introduction/Motivation

More information

IMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS

IMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS IMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS P.Mahalakshmi 1, J.Muthulakshmi 2, S.Kannadhasan 3 1,2 U.G Student, 3 Assistant Professor, Department of Electronics

More information

Adaptive Kernel Regression for Image Processing and Reconstruction

Adaptive Kernel Regression for Image Processing and Reconstruction Adaptive Kernel Regression for Image Processing and Reconstruction Peyman Milanfar* EE Department University of California, Santa Cruz *Joint work with Sina Farsiu, Hiro Takeda AFOSR Sensing Program Review,

More information

Locally Adaptive Learning for Translation-Variant MRF Image Priors

Locally Adaptive Learning for Translation-Variant MRF Image Priors Locally Adaptive Learning for Translation-Variant MRF Image Priors Masayuki Tanaka and Masatoshi Okutomi Tokyo Institute of Technology 2-12-1 O-okayama, Meguro-ku, Tokyo, JAPAN mtanaka@ok.ctrl.titech.ac.p,

More information

Image Restoration Using DNN

Image Restoration Using DNN Image Restoration Using DNN Hila Levi & Eran Amar Images were taken from: http://people.tuebingen.mpg.de/burger/neural_denoising/ Agenda Domain Expertise vs. End-to-End optimization Image Denoising and

More information

Blind Image Deblurring Using Dark Channel Prior

Blind Image Deblurring Using Dark Channel Prior Blind Image Deblurring Using Dark Channel Prior Jinshan Pan 1,2,3, Deqing Sun 2,4, Hanspeter Pfister 2, and Ming-Hsuan Yang 3 1 Dalian University of Technology 2 Harvard University 3 UC Merced 4 NVIDIA

More information

Introduction to patch-based approaches for image processing F. Tupin

Introduction to patch-based approaches for image processing F. Tupin Introduction to patch-based approaches for image processing F. Tupin Athens week Introduction Image processing Denoising and models Non-local / patch based approaches Principle Toy examples Limits and

More information

Image Denoising using Locally Learned Dictionaries

Image Denoising using Locally Learned Dictionaries Image Denoising using Locally Learned Dictionaries Priyam Chatterjee and Peyman Milanfar Department of Electrical Engineering, University of California, Santa Cruz, CA 95064, USA. ABSTRACT In this paper

More information

SUPPLEMENTARY MATERIAL

SUPPLEMENTARY MATERIAL SUPPLEMENTARY MATERIAL Zhiyuan Zha 1,3, Xin Liu 2, Ziheng Zhou 2, Xiaohua Huang 2, Jingang Shi 2, Zhenhong Shang 3, Lan Tang 1, Yechao Bai 1, Qiong Wang 1, Xinggan Zhang 1 1 School of Electronic Science

More information

Statistical image models

Statistical image models Chapter 4 Statistical image models 4. Introduction 4.. Visual worlds Figure 4. shows images that belong to different visual worlds. The first world (fig. 4..a) is the world of white noise. It is the world

More information

Computational Photography Denoising

Computational Photography Denoising Computational Photography Denoising Jongmin Baek CS 478 Lecture Feb 13, 2012 Announcements Term project proposal Due Wednesday Proposal presentation Next Wednesday Send us your slides (Keynote, PowerPoint,

More information

Generalized Tree-Based Wavelet Transform and Applications to Patch-Based Image Processing

Generalized Tree-Based Wavelet Transform and Applications to Patch-Based Image Processing Generalized Tree-Based Wavelet Transform and * Michael Elad The Computer Science Department The Technion Israel Institute of technology Haifa 32000, Israel *Joint work with A Seminar in the Hebrew University

More information

CLASS-SPECIFIC POISSON DENOISING BY PATCH-BASED IMPORTANCE SAMPLING

CLASS-SPECIFIC POISSON DENOISING BY PATCH-BASED IMPORTANCE SAMPLING CLASS-SPECIFIC POISSON DENOISING BY PATCH-BASED IMPORTANCE SAMPLING Milad Ninejad, José M. Bioucas-Dias, Mário A. T. Figueiredo Instituto de Telecomunicacoes Instituto Superior Tecnico Universidade de

More information

Patch-Based Color Image Denoising using efficient Pixel-Wise Weighting Techniques

Patch-Based Color Image Denoising using efficient Pixel-Wise Weighting Techniques Patch-Based Color Image Denoising using efficient Pixel-Wise Weighting Techniques Syed Gilani Pasha Assistant Professor, Dept. of ECE, School of Engineering, Central University of Karnataka, Gulbarga,

More information

Implementation of the Non-Local Bayes (NL-Bayes) Image Denoising Algorithm

Implementation of the Non-Local Bayes (NL-Bayes) Image Denoising Algorithm 2014/07/01 v0.5 IPOL article class Published in Image Processing On Line on 2013 06 17. Submitted on 2012 05 22, accepted on 2013 02 14. ISSN 2105 1232 c 2013 IPOL & the authors CC BY NC SA This article

More information

Locally Adaptive Regression Kernels with (many) Applications

Locally Adaptive Regression Kernels with (many) Applications Locally Adaptive Regression Kernels with (many) Applications Peyman Milanfar EE Department University of California, Santa Cruz Joint work with Hiro Takeda, Hae Jong Seo, Xiang Zhu Outline Introduction/Motivation

More information

Explore the Power of External Data in Denoising Task

Explore the Power of External Data in Denoising Task Explore the Power of External Data in Denoising Task Yipin Zhou Brown University Abstract The goal of this paper is to explore the power of external data in the image denoising task, that is, to show that

More information

One Network to Solve Them All Solving Linear Inverse Problems using Deep Projection Models

One Network to Solve Them All Solving Linear Inverse Problems using Deep Projection Models One Network to Solve Them All Solving Linear Inverse Problems using Deep Projection Models [Supplemental Materials] 1. Network Architecture b ref b ref +1 We now describe the architecture of the networks

More information

IMAGE DENOISING BY TARGETED EXTERNAL DATABASES

IMAGE DENOISING BY TARGETED EXTERNAL DATABASES IMAGE DENOISING BY TARGETED EXTERNAL DATABASES Enming Luo 1, Stanley H. Chan, and Truong Q. Nguyen 1 1 University of California, San Diego, Dept of ECE, 9500 Gilman Drive, La Jolla, CA 9093. Harvard School

More information

Graphical Models for Computer Vision

Graphical Models for Computer Vision Graphical Models for Computer Vision Pedro F Felzenszwalb Brown University Joint work with Dan Huttenlocher, Joshua Schwartz, Ross Girshick, David McAllester, Deva Ramanan, Allie Shapiro, John Oberlin

More information

arxiv: v1 [cs.cv] 2 Dec 2015

arxiv: v1 [cs.cv] 2 Dec 2015 MMSE Estimation for Poisson Noise Removal in Images arxiv:1512.00717v1 [cs.cv] 2 Dec 2015 Stanislav Pyatykh* and Jürgen Hesser December 3, 2015 Abstract Poisson noise suppression is an important preprocessing

More information

Available online at ScienceDirect. Procedia Computer Science 54 (2015 ) Mayank Tiwari and Bhupendra Gupta

Available online at   ScienceDirect. Procedia Computer Science 54 (2015 ) Mayank Tiwari and Bhupendra Gupta Available online at www.sciencedirect.com ScienceDirect Procedia Computer Science 54 (2015 ) 638 645 Eleventh International Multi-Conference on Information Processing-2015 (IMCIP-2015) Image Denoising

More information

Improved Non-Local Means Algorithm Based on Dimensionality Reduction

Improved Non-Local Means Algorithm Based on Dimensionality Reduction Improved Non-Local Means Algorithm Based on Dimensionality Reduction Golam M. Maruf and Mahmoud R. El-Sakka (&) Department of Computer Science, University of Western Ontario, London, Ontario, Canada {gmaruf,melsakka}@uwo.ca

More information

A non-local algorithm for image denoising

A non-local algorithm for image denoising A non-local algorithm for image denoising Antoni Buades, Bartomeu Coll Dpt. Matemàtiques i Informàtica, UIB Ctra. Valldemossa Km. 7.5, 07122 Palma de Mallorca, Spain vdmiabc4@uib.es, tomeu.coll@uib.es

More information

Image pyramids and their applications Bill Freeman and Fredo Durand Feb. 28, 2006

Image pyramids and their applications Bill Freeman and Fredo Durand Feb. 28, 2006 Image pyramids and their applications 6.882 Bill Freeman and Fredo Durand Feb. 28, 2006 Image pyramids Gaussian Laplacian Wavelet/QMF Steerable pyramid http://www-bcs.mit.edu/people/adelson/pub_pdfs/pyramid83.pdf

More information

An Optimized Pixel-Wise Weighting Approach For Patch-Based Image Denoising

An Optimized Pixel-Wise Weighting Approach For Patch-Based Image Denoising An Optimized Pixel-Wise Weighting Approach For Patch-Based Image Denoising Dr. B. R.VIKRAM M.E.,Ph.D.,MIEEE.,LMISTE, Principal of Vijay Rural Engineering College, NIZAMABAD ( Dt.) G. Chaitanya M.Tech,

More information

08 An Introduction to Dense Continuous Robotic Mapping

08 An Introduction to Dense Continuous Robotic Mapping NAVARCH/EECS 568, ROB 530 - Winter 2018 08 An Introduction to Dense Continuous Robotic Mapping Maani Ghaffari March 14, 2018 Previously: Occupancy Grid Maps Pose SLAM graph and its associated dense occupancy

More information

Expected Patch Log Likelihood with a Sparse Prior

Expected Patch Log Likelihood with a Sparse Prior Expected Patch Log Likelihood with a Sparse Prior Jeremias Sulam and Michael Elad Computer Science Department, Technion, Israel {jsulam,elad}@cs.technion.ac.il Abstract. Image priors are of great importance

More information

Filters. Advanced and Special Topics: Filters. Filters

Filters. Advanced and Special Topics: Filters. Filters Filters Advanced and Special Topics: Filters Dr. Edmund Lam Department of Electrical and Electronic Engineering The University of Hong Kong ELEC4245: Digital Image Processing (Second Semester, 2016 17)

More information

A Comparative Study & Analysis of Image Restoration by Non Blind Technique

A Comparative Study & Analysis of Image Restoration by Non Blind Technique A Comparative Study & Analysis of Image Restoration by Non Blind Technique Saurav Rawat 1, S.N.Tazi 2 M.Tech Student, Assistant Professor, CSE Department, Government Engineering College, Ajmer Abstract:

More information

AN ALGORITHM FOR BLIND RESTORATION OF BLURRED AND NOISY IMAGES

AN ALGORITHM FOR BLIND RESTORATION OF BLURRED AND NOISY IMAGES AN ALGORITHM FOR BLIND RESTORATION OF BLURRED AND NOISY IMAGES Nader Moayeri and Konstantinos Konstantinides Hewlett-Packard Laboratories 1501 Page Mill Road Palo Alto, CA 94304-1120 moayeri,konstant@hpl.hp.com

More information

Instance-based Learning CE-717: Machine Learning Sharif University of Technology. M. Soleymani Fall 2015

Instance-based Learning CE-717: Machine Learning Sharif University of Technology. M. Soleymani Fall 2015 Instance-based Learning CE-717: Machine Learning Sharif University of Technology M. Soleymani Fall 2015 Outline Non-parametric approach Unsupervised: Non-parametric density estimation Parzen Windows K-Nearest

More information

Multi-frame super-resolution with no explicit motion estimation

Multi-frame super-resolution with no explicit motion estimation Multi-frame super-resolution with no explicit motion estimation Mehran Ebrahimi and Edward R. Vrscay Department of Applied Mathematics Faculty of Mathematics, University of Waterloo Waterloo, Ontario,

More information

Bayesian Spherical Wavelet Shrinkage: Applications to Shape Analysis

Bayesian Spherical Wavelet Shrinkage: Applications to Shape Analysis Bayesian Spherical Wavelet Shrinkage: Applications to Shape Analysis Xavier Le Faucheur a, Brani Vidakovic b and Allen Tannenbaum a a School of Electrical and Computer Engineering, b Department of Biomedical

More information

When Sparsity Meets Low-Rankness: Transform Learning With Non-Local Low-Rank Constraint For Image Restoration

When Sparsity Meets Low-Rankness: Transform Learning With Non-Local Low-Rank Constraint For Image Restoration When Sparsity Meets Low-Rankness: Transform Learning With Non-Local Low-Rank Constraint For Image Restoration Bihan Wen, Yanjun Li and Yoram Bresler Department of Electrical and Computer Engineering Coordinated

More information

Perception Viewed as an Inverse Problem

Perception Viewed as an Inverse Problem Perception Viewed as an Inverse Problem Fechnerian Causal Chain of Events - an evaluation Fechner s study of outer psychophysics assumes that the percept is a result of a causal chain of events: Distal

More information

Locally Weighted Least Squares Regression for Image Denoising, Reconstruction and Up-sampling

Locally Weighted Least Squares Regression for Image Denoising, Reconstruction and Up-sampling Locally Weighted Least Squares Regression for Image Denoising, Reconstruction and Up-sampling Moritz Baecher May 15, 29 1 Introduction Edge-preserving smoothing and super-resolution are classic and important

More information

Image Redundancy and Non-Parametric Estimation for Image Representation

Image Redundancy and Non-Parametric Estimation for Image Representation Image Redundancy and Non-Parametric Estimation for Image Representation Charles Kervrann, Patrick Pérez, Jérôme Boulanger INRIA Rennes / INRA MIA Jouy-en-Josas / Institut Curie Paris VISTA Team http://www.irisa.fr/vista/

More information

Synthesis and Analysis Sparse Representation Models for Image Restoration. Shuhang Gu 顾舒航. Dept. of Computing The Hong Kong Polytechnic University

Synthesis and Analysis Sparse Representation Models for Image Restoration. Shuhang Gu 顾舒航. Dept. of Computing The Hong Kong Polytechnic University Synthesis and Analysis Sparse Representation Models for Image Restoration Shuhang Gu 顾舒航 Dept. of Computing The Hong Kong Polytechnic University Outline Sparse representation models for image modeling

More information

Image Processing with Nonparametric Neighborhood Statistics

Image Processing with Nonparametric Neighborhood Statistics Image Processing with Nonparametric Neighborhood Statistics Ross T. Whitaker Scientific Computing and Imaging Institute School of Computing University of Utah PhD Suyash P. Awate University of Pennsylvania,

More information

IMAGE DENOISING BY TARGETED EXTERNAL DATABASES

IMAGE DENOISING BY TARGETED EXTERNAL DATABASES 014 IEEE International Conference on Acoustic, Speech and Signal Processing (ICASSP) IMAGE DENOISING BY TARGETED EXTERNAL DATABASES Enming Luo 1, Stanley H. Chan, and Truong Q. Nguyen 1 1 University of

More information

FMA901F: Machine Learning Lecture 3: Linear Models for Regression. Cristian Sminchisescu

FMA901F: Machine Learning Lecture 3: Linear Models for Regression. Cristian Sminchisescu FMA901F: Machine Learning Lecture 3: Linear Models for Regression Cristian Sminchisescu Machine Learning: Frequentist vs. Bayesian In the frequentist setting, we seek a fixed parameter (vector), with value(s)

More information

Image Restoration with Deep Generative Models

Image Restoration with Deep Generative Models Image Restoration with Deep Generative Models Raymond A. Yeh *, Teck-Yian Lim *, Chen Chen, Alexander G. Schwing, Mark Hasegawa-Johnson, Minh N. Do Department of Electrical and Computer Engineering, University

More information

GRID WARPING IN TOTAL VARIATION IMAGE ENHANCEMENT METHODS. Andrey Nasonov, and Andrey Krylov

GRID WARPING IN TOTAL VARIATION IMAGE ENHANCEMENT METHODS. Andrey Nasonov, and Andrey Krylov GRID WARPING IN TOTAL VARIATION IMAGE ENHANCEMENT METHODS Andrey Nasonov, and Andrey Krylov Lomonosov Moscow State University, Moscow, Department of Computational Mathematics and Cybernetics, e-mail: nasonov@cs.msu.ru,

More information

Use of Extreme Value Statistics in Modeling Biometric Systems

Use of Extreme Value Statistics in Modeling Biometric Systems Use of Extreme Value Statistics in Modeling Biometric Systems Similarity Scores Two types of matching: Genuine sample Imposter sample Matching scores Enrolled sample 0.95 0.32 Probability Density Decision

More information

The Trainable Alternating Gradient Shrinkage method

The Trainable Alternating Gradient Shrinkage method The Trainable Alternating Gradient Shrinkage method Carlos Malanche, supervised by Ha Nguyen April 24, 2018 LIB (prof. Michael Unser), EPFL Quick example 1 The reconstruction problem Challenges in image

More information

Machine Learning / Jan 27, 2010

Machine Learning / Jan 27, 2010 Revisiting Logistic Regression & Naïve Bayes Aarti Singh Machine Learning 10-701/15-781 Jan 27, 2010 Generative and Discriminative Classifiers Training classifiers involves learning a mapping f: X -> Y,

More information

Bilevel Sparse Coding

Bilevel Sparse Coding Adobe Research 345 Park Ave, San Jose, CA Mar 15, 2013 Outline 1 2 The learning model The learning algorithm 3 4 Sparse Modeling Many types of sensory data, e.g., images and audio, are in high-dimensional

More information

Spatio-Temporal Markov Random Field for Video Denoising

Spatio-Temporal Markov Random Field for Video Denoising Spatio-Temporal Markov Random Field for Video Denoising Jia Chen Chi-Keung Tang Vision and Graphics Group The Hong Kong University of Science and Technology {jiachen,cktang}@cse.ust.hk Abstract This paper

More information

A Fast Non-Local Image Denoising Algorithm

A Fast Non-Local Image Denoising Algorithm A Fast Non-Local Image Denoising Algorithm A. Dauwe, B. Goossens, H.Q. Luong and W. Philips IPI-TELIN-IBBT, Ghent University, Sint-Pietersnieuwstraat 41, 9000 Ghent, Belgium ABSTRACT In this paper we propose

More information

CHAPTER VIII SEGMENTATION USING REGION GROWING AND THRESHOLDING ALGORITHM

CHAPTER VIII SEGMENTATION USING REGION GROWING AND THRESHOLDING ALGORITHM CHAPTER VIII SEGMENTATION USING REGION GROWING AND THRESHOLDING ALGORITHM 8.1 Algorithm Requirement The analysis of medical images often requires segmentation prior to visualization or quantification.

More information

Scene Grammars, Factor Graphs, and Belief Propagation

Scene Grammars, Factor Graphs, and Belief Propagation Scene Grammars, Factor Graphs, and Belief Propagation Pedro Felzenszwalb Brown University Joint work with Jeroen Chua Probabilistic Scene Grammars General purpose framework for image understanding and

More information

Analysis of Various Issues in Non-Local Means Image Denoising Algorithm

Analysis of Various Issues in Non-Local Means Image Denoising Algorithm Analysis of Various Issues in Non-Local Means Image Denoising Algorithm Sonika 1, Shalini Aggarwal 2, Pardeep kumar 3 Department of Computer Science and Applications, Kurukshetra University, Kurukshetra,

More information

Investigating Image Inpainting via the Alternating Direction Method of Multipliers

Investigating Image Inpainting via the Alternating Direction Method of Multipliers Investigating Image Inpainting via the Alternating Direction Method of Multipliers Jonathan Tuck Stanford University jonathantuck@stanford.edu Abstract In many imaging applications, there exists potential

More information

Face Hallucination Based on Eigentransformation Learning

Face Hallucination Based on Eigentransformation Learning Advanced Science and Technology etters, pp.32-37 http://dx.doi.org/10.14257/astl.2016. Face allucination Based on Eigentransformation earning Guohua Zou School of software, East China University of Technology,

More information

Image Processing using Smooth Ordering of its Patches

Image Processing using Smooth Ordering of its Patches 1 Image Processing using Smooth Ordering of its Patches Idan Ram, Michael Elad, Fellow, IEEE, and Israel Cohen, Senior Member, IEEE arxiv:12.3832v1 [cs.cv] 14 Oct 212 Abstract We propose an image processing

More information

MULTIFRAME RAW-DATA DENOISING BASED ON BLOCK-MATCHING AND 3-D D FILTERING FOR LOW-LIGHT LIGHT IMAGING AND STABILIZATION

MULTIFRAME RAW-DATA DENOISING BASED ON BLOCK-MATCHING AND 3-D D FILTERING FOR LOW-LIGHT LIGHT IMAGING AND STABILIZATION MULTIFRAME RAW-DATA DENOISING BASED ON BLOCK-MATCHING AND 3-D D FILTERING FOR LOW-LIGHT LIGHT IMAGING AND STABILIZATION LNLA 2008 - The 2008 International Workshop on Local and Non-Local Approximation

More information

Blind Deblurring using Internal Patch Recurrence. Tomer Michaeli & Michal Irani Weizmann Institute

Blind Deblurring using Internal Patch Recurrence. Tomer Michaeli & Michal Irani Weizmann Institute Blind Deblurring using Internal Patch Recurrence Tomer Michaeli & Michal Irani Weizmann Institute Scale Invariance in Natural Images Small patterns recur at different scales Scale Invariance in Natural

More information

Exploring Curve Fitting for Fingers in Egocentric Images

Exploring Curve Fitting for Fingers in Egocentric Images Exploring Curve Fitting for Fingers in Egocentric Images Akanksha Saran Robotics Institute, Carnegie Mellon University 16-811: Math Fundamentals for Robotics Final Project Report Email: asaran@andrew.cmu.edu

More information

Scene Grammars, Factor Graphs, and Belief Propagation

Scene Grammars, Factor Graphs, and Belief Propagation Scene Grammars, Factor Graphs, and Belief Propagation Pedro Felzenszwalb Brown University Joint work with Jeroen Chua Probabilistic Scene Grammars General purpose framework for image understanding and

More information

K-means and Hierarchical Clustering

K-means and Hierarchical Clustering K-means and Hierarchical Clustering Xiaohui Xie University of California, Irvine K-means and Hierarchical Clustering p.1/18 Clustering Given n data points X = {x 1, x 2,, x n }. Clustering is the partitioning

More information

Note Set 4: Finite Mixture Models and the EM Algorithm

Note Set 4: Finite Mixture Models and the EM Algorithm Note Set 4: Finite Mixture Models and the EM Algorithm Padhraic Smyth, Department of Computer Science University of California, Irvine Finite Mixture Models A finite mixture model with K components, for

More information

Bayesian Methods in Vision: MAP Estimation, MRFs, Optimization

Bayesian Methods in Vision: MAP Estimation, MRFs, Optimization Bayesian Methods in Vision: MAP Estimation, MRFs, Optimization CS 650: Computer Vision Bryan S. Morse Optimization Approaches to Vision / Image Processing Recurring theme: Cast vision problem as an optimization

More information

Image Restoration and Background Separation Using Sparse Representation Framework

Image Restoration and Background Separation Using Sparse Representation Framework Image Restoration and Background Separation Using Sparse Representation Framework Liu, Shikun Abstract In this paper, we introduce patch-based PCA denoising and k-svd dictionary learning method for the

More information

Using the Forest to See the Trees: Context-based Object Recognition

Using the Forest to See the Trees: Context-based Object Recognition Using the Forest to See the Trees: Context-based Object Recognition Bill Freeman Joint work with Antonio Torralba and Kevin Murphy Computer Science and Artificial Intelligence Laboratory MIT A computer

More information

Image Denoising Based on Hybrid Fourier and Neighborhood Wavelet Coefficients Jun Cheng, Songli Lei

Image Denoising Based on Hybrid Fourier and Neighborhood Wavelet Coefficients Jun Cheng, Songli Lei Image Denoising Based on Hybrid Fourier and Neighborhood Wavelet Coefficients Jun Cheng, Songli Lei College of Physical and Information Science, Hunan Normal University, Changsha, China Hunan Art Professional

More information

Convex Optimization MLSS 2015

Convex Optimization MLSS 2015 Convex Optimization MLSS 2015 Constantine Caramanis The University of Texas at Austin The Optimization Problem minimize : f (x) subject to : x X. The Optimization Problem minimize : f (x) subject to :

More information

Large Scale Data Analysis Using Deep Learning

Large Scale Data Analysis Using Deep Learning Large Scale Data Analysis Using Deep Learning Machine Learning Basics - 1 U Kang Seoul National University U Kang 1 In This Lecture Overview of Machine Learning Capacity, overfitting, and underfitting

More information

BM3D Image Denoising using Learning-based Adaptive Hard Thresholding

BM3D Image Denoising using Learning-based Adaptive Hard Thresholding BM3D Image Denoising using Learning-based Adaptive Hard Thresholding Farhan Bashar and Mahmoud R. El-Sakka Computer Science Department, The University of Western Ontario, London, Ontario, Canada Keywords:

More information

Missing Data Analysis for the Employee Dataset

Missing Data Analysis for the Employee Dataset Missing Data Analysis for the Employee Dataset 67% of the observations have missing values! Modeling Setup Random Variables: Y i =(Y i1,...,y ip ) 0 =(Y i,obs, Y i,miss ) 0 R i =(R i1,...,r ip ) 0 ( 1

More information

SEMI-BLIND IMAGE RESTORATION USING A LOCAL NEURAL APPROACH

SEMI-BLIND IMAGE RESTORATION USING A LOCAL NEURAL APPROACH SEMI-BLIND IMAGE RESTORATION USING A LOCAL NEURAL APPROACH Ignazio Gallo, Elisabetta Binaghi and Mario Raspanti Universitá degli Studi dell Insubria Varese, Italy email: ignazio.gallo@uninsubria.it ABSTRACT

More information

External Patch Prior Guided Internal Clustering for Image Denoising

External Patch Prior Guided Internal Clustering for Image Denoising External Patch Prior Guided Internal Clustering for Image Denoising Fei Chen 1, Lei Zhang 2, and Huimin Yu 3 1 College of Mathematics and Computer Science, Fuzhou University, Fuzhou, China 2 Dept. of Computing,

More information

How does bilateral filter relates with other methods?

How does bilateral filter relates with other methods? A Gentle Introduction to Bilateral Filtering and its Applications How does bilateral filter relates with other methods? Fredo Durand (MIT CSAIL) Slides by Pierre Kornprobst (INRIA) 0:35 Many people worked

More information

Detecting Salient Contours Using Orientation Energy Distribution. Part I: Thresholding Based on. Response Distribution

Detecting Salient Contours Using Orientation Energy Distribution. Part I: Thresholding Based on. Response Distribution Detecting Salient Contours Using Orientation Energy Distribution The Problem: How Does the Visual System Detect Salient Contours? CPSC 636 Slide12, Spring 212 Yoonsuck Choe Co-work with S. Sarma and H.-C.

More information

DS Machine Learning and Data Mining I. Alina Oprea Associate Professor, CCIS Northeastern University

DS Machine Learning and Data Mining I. Alina Oprea Associate Professor, CCIS Northeastern University DS 4400 Machine Learning and Data Mining I Alina Oprea Associate Professor, CCIS Northeastern University September 20 2018 Review Solution for multiple linear regression can be computed in closed form

More information

Clustering: Classic Methods and Modern Views

Clustering: Classic Methods and Modern Views Clustering: Classic Methods and Modern Views Marina Meilă University of Washington mmp@stat.washington.edu June 22, 2015 Lorentz Center Workshop on Clusters, Games and Axioms Outline Paradigms for clustering

More information

Hybrid Quasi-Monte Carlo Method for the Simulation of State Space Models

Hybrid Quasi-Monte Carlo Method for the Simulation of State Space Models The Tenth International Symposium on Operations Research and Its Applications (ISORA 211) Dunhuang, China, August 28 31, 211 Copyright 211 ORSC & APORC, pp. 83 88 Hybrid Quasi-Monte Carlo Method for the

More information

IMA Preprint Series # 2052

IMA Preprint Series # 2052 FAST IMAGE AND VIDEO DENOISING VIA NON-LOCAL MEANS OF SIMILAR NEIGHBORHOODS By Mona Mahmoudi and Guillermo Sapiro IMA Preprint Series # 2052 ( June 2005 ) INSTITUTE FOR MATHEMATICS AND ITS APPLICATIONS

More information

arxiv: v1 [cs.cv] 18 Jan 2019

arxiv: v1 [cs.cv] 18 Jan 2019 Good Similar Patches for Image Denoising Si Lu Portland State University lusi@pdx.edu arxiv:1901.06046v1 [cs.cv] 18 Jan 2019 Abstract Patch-based denoising algorithms like BM3D have achieved outstanding

More information

Divide and Conquer Kernel Ridge Regression

Divide and Conquer Kernel Ridge Regression Divide and Conquer Kernel Ridge Regression Yuchen Zhang John Duchi Martin Wainwright University of California, Berkeley COLT 2013 Yuchen Zhang (UC Berkeley) Divide and Conquer KRR COLT 2013 1 / 15 Problem

More information

arxiv: v1 [cs.cv] 30 Oct 2018

arxiv: v1 [cs.cv] 30 Oct 2018 Image Restoration using Total Variation Regularized Deep Image Prior Jiaming Liu 1, Yu Sun 2, Xiaojian Xu 2 and Ulugbek S. Kamilov 1,2 1 Department of Electrical and Systems Engineering, Washington University

More information

Image Denoising AGAIN!?

Image Denoising AGAIN!? 1 Image Denoising AGAIN!? 2 A Typical Imaging Pipeline 2 Sources of Noise (1) Shot Noise - Result of random photon arrival - Poisson distributed - Serious in low-light condition - Not so bad under good

More information

Image Processing Via Pixel Permutations

Image Processing Via Pixel Permutations Image Processing Via Pixel Permutations Michael Elad The Computer Science Department The Technion Israel Institute of technology Haifa 32000, Israel Joint work with Idan Ram Israel Cohen The Electrical

More information

Image Restoration Using Gaussian Mixture Models With Spatially Constrained Patch Clustering

Image Restoration Using Gaussian Mixture Models With Spatially Constrained Patch Clustering JOURNAL OF IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. XX, NO. X, 201X 1 Image Restoration Using Gaussian Mixture Models With Spatially Constrained Patch Clustering Milad Niknejad, Hossein Rabbani*, Senior

More information

DENOISING AN IMAGE BY DENOISING ITS CURVATURE IMAGE. Marcelo Bertalmío and Stacey Levine. IMA Preprint Series #2411.

DENOISING AN IMAGE BY DENOISING ITS CURVATURE IMAGE. Marcelo Bertalmío and Stacey Levine. IMA Preprint Series #2411. DENOISING AN IMAGE BY DENOISING ITS CURVATURE IMAGE By Marcelo Bertalmío and Stacey Levine IMA Preprint Series #2411 (September 2012) INSTITUTE FOR MATHEMATICS AND ITS APPLICATIONS UNIVERSITY OF MINNESOTA

More information

Image Denoising via Group Sparse Eigenvectors of Graph Laplacian

Image Denoising via Group Sparse Eigenvectors of Graph Laplacian Image Denoising via Group Sparse Eigenvectors of Graph Laplacian Yibin Tang, Ying Chen, Ning Xu, Aimin Jiang, Lin Zhou College of IOT Engineering, Hohai University, Changzhou, China School of Information

More information

Chapter 3. Bootstrap. 3.1 Introduction. 3.2 The general idea

Chapter 3. Bootstrap. 3.1 Introduction. 3.2 The general idea Chapter 3 Bootstrap 3.1 Introduction The estimation of parameters in probability distributions is a basic problem in statistics that one tends to encounter already during the very first course on the subject.

More information

Robert Collins CSE598G. Robert Collins CSE598G

Robert Collins CSE598G. Robert Collins CSE598G Recall: Kernel Density Estimation Given a set of data samples x i ; i=1...n Convolve with a kernel function H to generate a smooth function f(x) Equivalent to superposition of multiple kernels centered

More information

All images are degraded

All images are degraded Lecture 7 Image Relaxation: Restoration and Feature Extraction ch. 6 of Machine Vision by Wesley E. Snyder & Hairong Qi Spring 2018 16-725 (CMU RI) : BioE 2630 (Pitt) Dr. John Galeotti The content of these

More information

Digital Image Processing Laboratory: MAP Image Restoration

Digital Image Processing Laboratory: MAP Image Restoration Purdue University: Digital Image Processing Laboratories 1 Digital Image Processing Laboratory: MAP Image Restoration October, 015 1 Introduction This laboratory explores the use of maximum a posteriori

More information

ICA mixture models for image processing

ICA mixture models for image processing I999 6th Joint Sy~nposiurn orz Neural Computation Proceedings ICA mixture models for image processing Te-Won Lee Michael S. Lewicki The Salk Institute, CNL Carnegie Mellon University, CS & CNBC 10010 N.

More information

Learning to Perceive Transparency from the Statistics of Natural Scenes

Learning to Perceive Transparency from the Statistics of Natural Scenes Learning to Perceive Transparency from the Statistics of Natural Scenes Anat Levin Assaf Zomet Yair Weiss School of Computer Science and Engineering The Hebrew University of Jerusalem 9194 Jerusalem, Israel

More information

Structural Similarity Optimized Wiener Filter: A Way to Fight Image Noise

Structural Similarity Optimized Wiener Filter: A Way to Fight Image Noise Structural Similarity Optimized Wiener Filter: A Way to Fight Image Noise Mahmud Hasan and Mahmoud R. El-Sakka (B) Department of Computer Science, University of Western Ontario, London, ON, Canada {mhasan62,melsakka}@uwo.ca

More information

Multi-scale Statistical Image Models and Denoising

Multi-scale Statistical Image Models and Denoising Multi-scale Statistical Image Models and Denoising Eero P. Simoncelli Center for Neural Science, and Courant Institute of Mathematical Sciences New York University http://www.cns.nyu.edu/~eero Multi-scale

More information