What are Good Apertures for Defocus Deblurring? Columbia University ICCP 2009, San Francisco Changyin Zhou Shree K. Nayar.

Slides:



Advertisements
Similar presentations
Bayesian Belief Propagation
Advertisements

High frequency annular arrays can provide very good resolution in ultrasound bio-microscopy Numerical simulation of their imaging patterns is essential.
Range from Focus Based on the work of Eric Krotkov And Jean Paul Martin.
Image Registration  Mapping of Evolution. Registration Goals Assume the correspondences are known Find such f() and g() such that the images are best.
S INGLE -I MAGE R EFOCUSING AND D EFOCUSING Wei Zhang, Nember, IEEE, and Wai-Kuen Cham, Senior Member, IEEE.
Object Specific Compressed Sensing by minimizing a weighted L2-norm A. Mahalanobis.
--- some recent progress Bo Fu University of Kentucky.
Digital Image Processing, 2nd ed. © 2002 R. C. Gonzalez & R. E. Woods Chapter 5 Image Restoration Chapter 5 Image Restoration.
Light Fields PROPERTIES AND APPLICATIONS. Outline  What are light fields  Acquisition of light fields  from a 3D scene  from a real world scene 
Recap from Monday Frequency domain analytical tool computational shortcut compression tool.
Shaojie Zhuo, Dong Guo, Terence Sim School of Computing, National University of Singapore CVPR2010 Reporter: 周 澄 (A.J.) 01/16/2011 Key words: image deblur,
Unnatural L 0 Representation for Natural Image Deblurring Speaker: Wei-Sheng Lai Date: 2013/04/26.
Presenter: Yufan Liu November 17th,

ECE 472/572 - Digital Image Processing Lecture 8 - Image Restoration – Linear, Position-Invariant Degradations 10/10/11.
Optical Imaging in Astronomy 1st CASSDA School for Observers Observatorio del Teide, 20 – 25 April 2015 Franz Kneer Institut für Astrophysik Göttingen.
Diffusion Coding Photography for Extended Depth of Field SIGGRAPH 2010 Ollie Cossairt, Changyin Zhou, Shree Nayar Columbia University.
When Does a Camera See Rain? Department of Computer Science Columbia University Kshitiz Garg Shree K. Nayar ICCV Conference October 2005, Beijing, China.
Digital Image Processing Chapter 5: Image Restoration.
Project Presentation: March 9, 2006
CCU VISION LABORATORY Object Speed Measurements Using Motion Blurred Images 林惠勇 中正大學電機系
Interactive Matting Christoph Rhemann Supervised by: Margrit Gelautz and Carsten Rother.
Image Deblurring with Optimizations Qi Shan Leo Jiaya Jia Aseem Agarwala University of Washington The Chinese University of Hong Kong Adobe Systems, Inc.
Linear View Synthesis Using a Dimensionality Gap Light Field Prior
Design of Curves and Surfaces by Multi Objective Optimization Rony Goldenthal Michel Bercovier School of Computer Science and Engineering The Hebrew University.
1 Diffusion Coded Photography for Extended Depth of Field SIGGRAPH 2010 Oliver Cossairt, Changyin Zhou, Shree Nayar Columbia University Supported by ONR.
Lensless Imaging with A Controllable Aperture Assaf Zomet and Shree K. Nayar Columbia University IEEE CVPR Conference June 2006, New York, USA.
DIGITAL IMAGE PROCESSING Instructors: Dr J. Shanbehzadeh M.Gholizadeh M.Gholizadeh
Image deblurring Seminar inverse problems April 18th 2007 Willem Dijkstra.
Image Pyramids and Blending
Multi-Aperture Photography Paul Green – MIT CSAIL Wenyang Sun – MERL Wojciech Matusik – MERL Frédo Durand – MIT CSAIL.
Depth from Diffusion Supported by ONR Changyin ZhouShree NayarOliver Cossairt Columbia University.
1 Fabricating BRDFs at High Spatial Resolution Using Wave Optics Anat Levin, Daniel Glasner, Ying Xiong, Fredo Durand, Bill Freeman, Wojciech Matusik,
ELE 488 F06 ELE 488 Fall 2006 Image Processing and Transmission ( ) Wiener Filtering Derivation Comments Re-sampling and Re-sizing 1D  2D 10/5/06.
Introduction to Computational Photography. Computational Photography Digital Camera What is Computational Photography? Second breakthrough by IT First.
Shape Matching for Model Alignment 3D Scan Matching and Registration, Part I ICCV 2005 Short Course Michael Kazhdan Johns Hopkins University.
EE369C Final Project: Accelerated Flip Angle Sequences Jan 9, 2012 Jason Su.
EE4328, Section 005 Introduction to Digital Image Processing Linear Image Restoration Zhou Wang Dept. of Electrical Engineering The Univ. of Texas.
Extracting Barcodes from a Camera-Shaken Image on Camera Phones Graduate Institute of Communication Engineering National Taiwan University Chung-Hua Chu,
Yu-Wing Tai, Hao Du, Michael S. Brown, Stephen Lin CVPR’08 (Longer Version in Revision at IEEE Trans PAMI) Google Search: Video Deblurring Spatially Varying.
Reporter: Wade Chang Advisor: Jian-Jiun Ding 1 Depth Estimation and Focus Recovery.
1 Finding depth. 2 Overview Depth from stereo Depth from structured light Depth from focus / defocus Laser rangefinders.
Tutorial on Computational Optical Imaging University of Minnesota September David J. Brady Duke University
A Frequency-Domain Approach to Registration Estimation in 3-D Space Phillip Curtis Pierre Payeur Vision, Imaging, Video and Autonomous Systems Research.
Motion Deblurring Using Hybrid Imaging Moshe Ben-Ezra and Shree K. Nayar Columbia University IEEE CVPR Conference June 2003, Madison, USA.
CS654: Digital Image Analysis
Mitsubishi Electric Research Labs (MERL) Super-Res from Single Motion Blur PhotoAgrawal & Raskar Amit Agrawal and Ramesh Raskar Mitsubishi Electric Research.
The Media Lab Designing Aperture Masks in Phase Space Roarke Horstmeyer 1, Se Baek Oh 2, and Ramesh Raskar 1 1 MIT Media Lab 2 MIT Dept. of Mechanical.
CS654: Digital Image Analysis Lecture 22: Image Restoration - II.
Extracting Depth and Matte using a Color-Filtered Aperture Yosuke Bando TOSHIBA + The University of Tokyo Bing-Yu Chen National Taiwan University Tomoyuki.
1 Motion Blur Identification in Noisy Images Using Fuzzy Sets IEEE 5th International Symposium on Signal Processing and Information Technology (ISSPIT.
On the Evaluation of Optical Performace of Observing Instruments Y. Suematsu (National Astronomical Observatory of Japan) ABSTRACT: It is useful to represent.
Page 1© Crown copyright 2004 The use of an intensity-scale technique for assessing operational mesoscale precipitation forecasts Marion Mittermaier and.
Learning Photographic Global Tonal Adjustment with a Database of Input / Output Image Pairs.
Vincent DeVito Computer Systems Lab The goal of my project is to take an image input, artificially blur it using a known blur kernel, then.
For off-center points on screen, Fresnel zones on aperture are displaced …harder to “integrate” mentally. When white and black areas are equal, light at.
Multiresolution Histograms and their Use for Texture Classification Stathis Hadjidemetriou, Michael Grossberg and Shree Nayar CAVE Lab, Columbia University.
Today Defocus Deconvolution / inverse filters. Defocus.
ICCV 2007 National Laboratory of Pattern Recognition Institute of Automation Chinese Academy of Sciences Half Quadratic Analysis for Mean Shift: with Extension.
Noise Filtering in Monte Carlo Rendering
Zachary Starr Dept. of Computer Science, University of Missouri, Columbia, MO 65211, USA Digital Image Processing Final Project Dec 11 th /16 th, 2014.
Extended Depth of Field For Long Distance Biometrics
Image Deblurring and noise reduction in python
Deconvolution , , Computational Photography
Rob Fergus Computer Vision
Solving an estimation problem
Fourier Optics P47 – Optics: Unit 8.
Applications of Fourier Analysis in Image Recovery
ECE 299 Holography and Coherent Imaging
Deblurring Shaken and Partially Saturated Images
Presentation transcript:

What are Good Apertures for Defocus Deblurring? Columbia University ICCP 2009, San Francisco Changyin Zhou Shree K. Nayar

Geometry of Defocus Lens Object Sensor Focus Plane Aperture Pattern PSF

Formulation of Defocus Focused Image PSFImage Noise Captured Image In the Spatial Domain

Formulation of Defocus Focused Image PSFImage Noise Captured Image In the Fourier Domain

Various Aperture Coding Techniques [Veeraraghavan et al. 2007] [Levin et al. 2007] [Welford 1960] Annular Pattern [Gottesman 1989] MURA Pattern [Dowski 1993] Wave-front Coding

What are Good Apertures for Defocus Deblurring?

Outline - Evaluation Criterion for Aperture Patterns - Pattern Optimization - Background - Experiments

How to Evaluate Aperture Patterns? Focused Image Captured Image An Optimal Deblurring Algorithm Deblurred Image Natural image Gaussian white noise Evaluate K using expected quality of deblurred image Basic idea:

How to Evaluate Aperture Patterns? Focused Image Captured Image An Optimal Deblurring Algorithm Deblurred Image Natural image Gaussian white noise Linear system

The Optimal Linear Deblurring Algorithm For white noise and L 2 distance, Weiner filter is optimal Optimize C by minimizing the expected recovery error

The Optimal Linear Deblurring Algorithm A variant of Weiner deconvolution 1/f law : Expected power spectrum of natural images : Gaussian noise level

How to Evaluate Aperture Patterns? Focused Image Captured Image An Optimal Deblurring Algorithm Deblurred Image Natural image Gaussian white noise Linear system Evaluation Function:

Evaluation Criterion for Aperture Pattern Expected deblurring quality at frequency R(K | ) : Expected deblurring quality at noise level. Noise Level Frequency Prior Power Spectrum

Noise Level Evaluate Patterns Using the Criterion R Curve of Circular Pattern Noise Level R

Evaluate Patterns Using the Criterion Coded apertures help more when noise level is low. Circular aperture is better when noise level is high. Curves of Relative R: R(K)/R(K 0 ) Noise Level Veeraraghavan Levin Annular Image(binary) Image(gray) Random

Outline - Evaluation Criterion for Aperture Patterns - Pattern Optimization - Background - Experiments

Pattern Optimization Difficult to solve analytically Difficult to solve analytically Patterns evaluated in the Fourier domain, but strictly constrained in the spatial domain. Patterns evaluated in the Fourier domain, but strictly constrained in the spatial domain. Difficult to do brute force search Difficult to do brute force search For binary patterns of resolution N x N, the number of possible solutions is huge, For binary patterns of resolution N x N, the number of possible solutions is huge, when N = 13, if evaluating one pattern takes 1 millisecond, the brute force search requires yrs.

Pattern Optimization Pattern Evolution in Genetic Algorithm: 13 x 13 binary patterns; 8 different noise levels 2 nd Run: 3 nd Run: σ = σ =0.001 σ =0.002 σ =0.005 σ =0.008 σ =0.01 σ =0.02 σ =0.03

Evaluate the Optimized Patterns

Relative R curves of the Optimized Patterns CircularVeeraraghavan σ= Noise Level Image

Outline - Evaluation Criterion for Aperture Patterns - Pattern Optimization - Background - Experiments

Implementation Precision Laser Photoplot (1 micron)

Implementation Canon EF 50mm f/1.8 Lens

Implementation Image Pattern Levin et al.’s Pattern Our Optimized Pattern Circular Pattern (wide open) Veeraraghavan et al.’s Pattern

Comparison Experiments on a CZP Chart Focused Image Captured Images Circular Pattern Levin et al.’s Pattern Veeraraghavan et al’s PatternImage Pattern Our Optimized Pattern

Comparison Experiments on a CZP Chart Deblurred Images Focused ImageCircular PatternLevin et al.’s Pattern Veeraraghavan et al’s PatternImage Pattern Our Optimized Pattern

Shrek Captured Image using the Optimized Pattern

Shrek Captured Image using the Optimized Pattern Deblurring Result

On the Street Captured Image Using the Optimized Pattern

On the Street Captured Image Using the Optimized Pattern Deblurring Result

Traffic Scene Captured Image Using the Optimized Pattern

Traffic Scene Captured Image Using the Optimized Pattern Deblurring Result

Summary Main Contributions Aperture Evaluation Criterion for Defocus Deblurring Aperture Evaluation Criterion for Defocus Deblurring Aperture Pattern Optimization for Defocus Deblurring Aperture Pattern Optimization for Defocus Deblurring Future Work Optimize gray-level patterns Optimize gray-level patterns Apply Criterion to other PSF Engineering Problems Apply Criterion to other PSF Engineering Problems Extend Analysis to Account for Diffraction Extend Analysis to Account for Diffraction

Thank You!

Genetic Algorithm for Pattern Optimization

Noise level = st Run: G = 1G = 10G = 20G = 40G = 60G = 80 2 nd Run: 3 nd Run: G = 1G = 10G = 20G = 40G = 60G = 80 G = 1G = 10G = 20G = 40G = 60G = 80