Automatic Estimation and Removal of Noise from a Single Image

Slides:



Advertisements
Similar presentations
Bayesian Belief Propagation
Advertisements

S INGLE -I MAGE R EFOCUSING AND D EFOCUSING Wei Zhang, Nember, IEEE, and Wai-Kuen Cham, Senior Member, IEEE.
R. DOSIL, X. M. PARDO, A. MOSQUERA, D. CABELLO Grupo de Visión Artificial Departamento de Electrónica e Computación Universidade de Santiago de Compostela.
Analysis of Contour Motions Ce Liu William T. Freeman Edward H. Adelson Computer Science and Artificial Intelligence Laboratory Massachusetts Institute.
Topic 6 - Image Filtering - I DIGITAL IMAGE PROCESSING Course 3624 Department of Physics and Astronomy Professor Bob Warwick.
Digital Photography with Flash and No-Flash Image Pairs By: Georg PetschniggManeesh Agrawala Hugues HoppeRichard Szeliski Michael CohenKentaro Toyama,
Patch-based Image Deconvolution via Joint Modeling of Sparse Priors Chao Jia and Brian L. Evans The University of Texas at Austin 12 Sep
5. 1 Model of Image degradation and restoration
Image Denoising using Locally Learned Dictionaries Priyam Chatterjee Peyman Milanfar Dept. of Electrical Engineering University of California, Santa Cruz.
Foundations & Core in Computer Vision: A System Perspective Ce Liu Microsoft Research New England.
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.
Edge detection Goal: Identify sudden changes (discontinuities) in an image Intuitively, most semantic and shape information from the image can be encoded.
Wavelet-Domain Video Denoising Based on Reliability Measures Vladimir Zlokolica, Aleksandra Piˇzurica and Wilfried Philips Circuits and Systems for Video.
A New Block Based Motion Estimation with True Region Motion Field Jozef Huska & Peter Kulla EUROCON 2007 The International Conference on “Computer as a.
Retinex Algorithm Combined with Denoising Methods Hae Jong, Seo Multi Dimensional Signal Processing Group University of California at Santa Cruz.
Image processing. Image operations Operations on an image –Linear filtering –Non-linear filtering –Transformations –Noise removal –Segmentation.
Digital Image Processing Ligang Liu Zhejiang University
MSU CSE 803 Stockman Linear Operations Using Masks Masks are patterns used to define the weights used in averaging the neighbors of a pixel to compute.
High-Quality Video View Interpolation
SUSAN: structure-preserving noise reduction EE264: Image Processing Final Presentation by Luke Johnson 6/7/2007.
Texture Reading: Chapter 9 (skip 9.4) Key issue: How do we represent texture? Topics: –Texture segmentation –Texture-based matching –Texture synthesis.
MSU CSE 803 Linear Operations Using Masks Masks are patterns used to define the weights used in averaging the neighbors of a pixel to compute some result.
Computational Photography: Image Processing Jinxiang Chai.
1 Image Processing(IP) 1. Introduction 2. Digital Image Fundamentals 3. Image Enhancement in the spatial Domain 4. Image Enhancement in the Frequency Domain.
Retinex by Two Bilateral Filters Michael Elad The CS Department The Technion – Israel Institute of technology Haifa 32000, Israel Scale-Space 2005 The.
(1) A probability model respecting those covariance observations: Gaussian Maximum entropy probability distribution for a given covariance observation.
CS448f: Image Processing For Photography and Vision Denoising.
Noise Estimation from a Single Image Ce Liu William T. FreemanRichard Szeliski Sing Bing Kang.
Introduction --Classification Shape ContourRegion Structural Syntactic Graph Tree Model-driven Data-driven Perimeter Compactness Eccentricity.
CSCE 441: Computer Graphics Image Filtering Jinxiang Chai.
Machine Vision ENT 273 Image Filters Hema C.R. Lecture 5.
Adaptive Regularization of the NL-Means : Application to Image and Video Denoising IEEE TRANSACTION ON IMAGE PROCESSING , VOL , 23 , NO,8 , AUGUST 2014.
Stereo Matching Information Permeability For Stereo Matching – Cevahir Cigla and A.Aydın Alatan – Signal Processing: Image Communication, 2013 Radiometric.
1 Patch Complexity, Finite Pixel Correlations and Optimal Denoising Anat Levin, Boaz Nadler, Fredo Durand and Bill Freeman Weizmann Institute, MIT CSAIL.
Mutual Information-based Stereo Matching Combined with SIFT Descriptor in Log-chromaticity Color Space Yong Seok Heo, Kyoung Mu Lee, and Sang Uk Lee.
Prakash Chockalingam Clemson University Non-Rigid Multi-Modal Object Tracking Using Gaussian Mixture Models Committee Members Dr Stan Birchfield (chair)
Automatic Registration of Color Images to 3D Geometry Computer Graphics International 2009 Yunzhen Li and Kok-Lim Low School of Computing National University.
1. Introduction Motion Segmentation The Affine Motion Model Contour Extraction & Shape Estimation Recursive Shape Estimation & Motion Estimation Occlusion.
Image-based rendering Michael F. Cohen Microsoft Research.
University of Texas at Austin CS384G - Computer Graphics Fall 2010 Don Fussell Image processing.
Image Processing Jitendra Malik. Different kinds of images Radiance images, where a pixel value corresponds to the radiance from some point in the scene.
Why is computer vision difficult?
Image Enhancement [DVT final project]
December 9, 2014Computer Vision Lecture 23: Motion Analysis 1 Now we will talk about… Motion Analysis.
Machine Vision ENT 273 Image Filters Hema C.R. Lecture 5.
The 18th Meeting on Image Recognition and Understanding 2015/7/29 Depth Image Enhancement Using Local Tangent Plane Approximations Kiyoshi MatsuoYoshimitsu.
Practical Poissonian-Gaussian Noise Modeling and Fitting for Single- Image Raw-Data Reporter: 沈廷翰 陳奇業.
1 Markov random field: A brief introduction (2) Tzu-Cheng Jen Institute of Electronics, NCTU
Approximate Inference: Decomposition Methods with Applications to Computer Vision Kyomin Jung ( KAIST ) Joint work with Pushmeet Kohli (Microsoft Research)
Autonomous Robots Vision © Manfred Huber 2014.
Intelligent Vision Systems ENT 496 Image Filtering and Enhancement Hema C.R. Lecture 4.
Ch.8 Efficient Coding of Visual Scenes by Grouping and Segmentation Bayesian Brain Tai Sing Lee and Alan L. Yuille Heo, Min-Oh.
MIT AI Lab / LIDS Laboatory for Information and Decision Systems & Artificial Intelligence Laboratory Massachusetts Institute of Technology A Unified Multiresolution.
Bivariate Splines for Image Denoising*° *Grant Fiddyment University of Georgia, 2008 °Ming-Jun Lai Dept. of Mathematics University of Georgia.
Speaker Min-Koo Kang March 26, 2013 Depth Enhancement Technique by Sensor Fusion: MRF-based approach.
Instructor: Mircea Nicolescu Lecture 5 CS 485 / 685 Computer Vision.
Lecture 8: Edges and Feature Detection
Edge Preserving Spatially Varying Mixtures for Image Segmentation Giorgos Sfikas, Christophoros Nikou, Nikolaos Galatsanos (CVPR 2008) Presented by Lihan.
Introduction to Medical Imaging Week 6: Introduction to Medical Imaging Week 6: Denoising (part II) – Variational Methods and Evolutions Guy Gilboa Course.
Processing Images and Video for An Impressionist Effect Automatic production of “painterly” animations from video clips. Extending existing algorithms.
PDES in image processing
CS 4501: Introduction to Computer Vision Sparse Feature Detectors: Harris Corner, Difference of Gaussian Connelly Barnes Slides from Jason Lawrence, Fei.
Directional Multiscale Modeling of Images
IMAGE PROCESSING INTENSITY TRANSFORMATION AND SPATIAL FILTERING
Image filtering Hybrid Images, Oliva et al.,
PRAKASH CHOCKALINGAM, NALIN PRADEEP, AND STAN BIRCHFIELD
Analysis of Contour Motions
Linear Operations Using Masks
Adaptive Cooperative Systems Chapter 6 Markov Random Fields
Presentation transcript:

Automatic Estimation and Removal of Noise from a Single Image Accepted for publication by IEEE Transactions on Pattern Analysis and Machine Intelligence Automatic Estimation and Removal of Noise from a Single Image Ce Liu* Richard Szeliski† Sing Bing Kang† C. Lawrence Zitnick† William T. Freeman* *CSAIL, MIT †Microsoft Research

The criteria of image denoising Perceptually flat regions should be flat Image boundaries should be preserved (neither blurred or sharpened) Texture details should not be lost Global contrast should be preserved No artifacts should be generated

Denoising is hard… Different source and type of noise How strong is the noise? Locally, it is hard to distinguish Texture vs. noise Object boundary vs. structural noise Speed? Quality?

Related work Wavelets (coring, GSM) Anisotropic diffusion (PDEs) FRAME & FOE Bilateral filtering Nonlocal methods Conditional Random Fields

Segmentation-based piecewise smooth image models An image can be reconstructed by Over-segmentation (your favorite method) Affine reconstruction for each segment (RGB is an affine function of xy) Boundary blur estimation Piecewise smooth image reconstruction

Piecewise smooth reconstruction (a) Original image (b) Segmentation (c) Per-segment affine reconstruction (d) Affine reconstruction + boundary blur

Important properties of piecewise smooth image model Piecewise smooth image model is consistent with sparse image prior horizontal vertical Piecewise smooth reconstruction Filter responses

Important properties of piecewise smooth image model The color distribution per segment can be well approximated by a line segment Sorted eigenvalues in RGB space for each segment. The reddish picture indicates that the largest eigenvalue is significantly bigger than the second and third RGB values projected onto the eigenvector w.r.t. the largest eigenvalue. It is almost identical to the original image.

Important properties of piecewise smooth image model The standard deviation of residual per each segment is the upper bound of the noise level in that segment s I Brightness Residual std. dev. Assume brightness mean I is accurate estimate Standard deviation s is an over-estimate: (may contain signal) The lower envelope is the upper bound of noise level function (NLF)

Insert the CVPR slides on noise estimation

Segmentation-based noise reduction The noise level (std dev) for each segment, each RGB channel (from the inferred NLF) 0th-order model Assume the pixel is independent of the neighbors (of course this is not true) Bayesian MAP inference for each pixel (trivial)

The results of 0th order model

1st order Gaussian CRF Of course the neighboring pixels are dependent A Gaussian conditional random field (GCRF) is formed to incorporate Distance to noise Distance to signal Image-dependent regularization Inference by conjugate gradient method

Experimental results 17 images are randomly selected from Berkeley image segmentation database Compare to bilateral filtering, curvature preserving PDE and wavelet GSM.

We win the PSNR battle

Visual inspection 10% AWGN PDE Wavelet GSM Ours Original

Close-up view 10% AWGN PDE Wavelet GSM Ours Original

Visual inspection 10% AWGN PDE Wavelet GSM Ours Original

Close-up view 10% AWGN PDE Wavelet GSM Ours Original

Apply to real CCD noise Estimated NLF

Close-up view Since wavelet GSM only takes one constant noise level, it either over-smoothes (as s =15%) or under-smoothes (s =10%). We estimate signal-dependent noise level (NLF), and our approach handles both high and low noise level well. Estimated NLF

A very challenging example Real digital camera noise (low light, high ISO) Denoised by wavelet GSM Denoised by our model Estimated NLF

Conclusion Noise estimation and removal using one framework: piecewise smooth image model The chrominance component of the noise can be significantly reduced by simple projection Computer vision algorithms can be automated by estimating the peripheral parameters such as noise level, blur level, lighting, …