1 Resolving the Generalized Bas-Relief Ambiguity by Entropy Minimization Neil G. Alldrin Satya P. Mallick David J. Kriegman University of California, San.

Slides:



Advertisements
Similar presentations
1 Photometric Stereo Reconstruction Dr. Maria E. Angelopoulou.
Advertisements

Photometric Image Formation
The fundamental matrix F
Lecture 11: Two-view geometry
Invariants (continued).
Computer Vision Spring ,-685 Instructor: S. Narasimhan Wean 5403 T-R 3:00pm – 4:20pm Lecture #12.
Lecture 35: Photometric stereo CS4670 / 5670 : Computer Vision Noah Snavely.
Machine Vision Laboratory The University of the West of England Bristol, UK 2 nd September 2010 ICEBT Photometric Stereo in 3D Biometrics Gary A. Atkinson.
Computer Vision CS 776 Spring 2014 Photometric Stereo and Illumination Cones Prof. Alex Berg (Slide credits to many folks on individual slides)
Shadow Scanning [Bouguet 1999] Desk Lamp Camera Stick or pencil Desk The idea [Bouguet and Perona’98] J.-Y. Bouguet and P. Perona, “3D Photography on your.
Two-View Geometry CS Sastry and Yang
Lighting affects appearance. What is the Question ? (based on work of Basri and Jacobs, ICCV 2001) Given an object described by its normal at each.
Lecture 8: Stereo.
ECCV 2002 Removing Shadows From Images G. D. Finlayson 1, S.D. Hordley 1 & M.S. Drew 2 1 School of Information Systems, University of East Anglia, UK 2.
Lecture 23: Photometric Stereo CS4670/5760: Computer Vision Kavita Bala Scott Wehrwein.
AA II z f Surface Radiance and Image Irradiance Same solid angle Pinhole Camera Model.
Uncalibrated Geometry & Stratification Sastry and Yang
Visibility Subspaces: Uncalibrated Photometric Stereo with Shadows Kalyan Sunkavalli, Harvard University Joint work with Todd Zickler and Hanspeter Pfister.
Multiple-view Reconstruction from Points and Lines
Computer Vision Sources, shading and photometric stereo Marc Pollefeys COMP 256.
© 2003 by Davi GeigerComputer Vision October 2003 L1.1 Structure-from-EgoMotion (based on notes from David Jacobs, CS-Maryland) Determining the 3-D structure.
A Theory of Locally Low Dimensional Light Transport Dhruv Mahajan (Columbia University) Ira Kemelmacher-Shlizerman (Weizmann Institute) Ravi Ramamoorthi.
CSE473/573 – Stereo Correspondence
Photometric Stereo Merle Norman Cosmetics, Los Angeles Readings R. Woodham, Photometric Method for Determining Surface Orientation from Multiple Images.
Introduction of the intrinsic image. Intrinsic Images The method of Finlayson & Hordley ( 2001 ) Two assumptions 1. the camera ’ s sensors are sufficiently.
Algorithm Evaluation and Error Analysis class 7 Multiple View Geometry Comp Marc Pollefeys.
Photometric Stereo & Shape from Shading
Stockman MSU/CSE Math models 3D to 2D Affine transformations in 3D; Projections 3D to 2D; Derivation of camera matrix form.
Lecture II-2: Probability Review
Statistical Color Models (SCM) Kyungnam Kim. Contents Introduction Trivariate Gaussian model Chromaticity models –Fixed planar chromaticity models –Zhu.
Computer Vision Spring ,-685 Instructor: S. Narasimhan PH A18B T-R 10:30am – 11:50am Lecture #13.
1/20 Obtaining Shape from Scanning Electron Microscope Using Hopfield Neural Network Yuji Iwahori 1, Haruki Kawanaka 1, Shinji Fukui 2 and Kenji Funahashi.
Geometry and Algebra of Multiple Views
Computer Vision - A Modern Approach Set: Sources, shadows and shading Slides by D.A. Forsyth Sources and shading How bright (or what colour) are objects?
Brief Introduction to Geometry and Vision
1 Preview At least two views are required to access the depth of a scene point and in turn to reconstruct scene structure Multiple views can be obtained.
Projective cameras Motivation Elements of Projective Geometry Projective structure from motion Planches : –
Analysis of Lighting Effects Outline: The problem Lighting models Shape from shading Photometric stereo Harmonic analysis of lighting.
Y. Moses 11 Combining Photometric and Geometric Constraints Yael Moses IDC, Herzliya Joint work with Ilan Shimshoni and Michael Lindenbaum, the Technion.
Shape from Shading and Texture. Lambertian Reflectance Model Diffuse surfaces appear equally bright from all directionsDiffuse surfaces appear equally.
A Photometric Approach for Estimating Normals and Tangents Michael Holroyd University of Virginia Jason Lawrence University of Virginia Greg Humphreys.
Sources, Shadows, and Shading
Multiview Geometry and Stereopsis. Inputs: two images of a scene (taken from 2 viewpoints). Output: Depth map. Inputs: multiple images of a scene. Output:
EECS 274 Computer Vision Sources, Shadows, and Shading.
Lighting affects appearance. How do we represent light? (1) Ideal distant point source: - No cast shadows - Light distant - Three parameters - Example:
1 Formation et Analyse d’Images Session 7 Daniela Hall 25 November 2004.
Understanding the effect of lighting in images Ronen Basri.
The Quotient Image: Class-based Recognition and Synthesis Under Varying Illumination T. Riklin-Raviv and A. Shashua Institute of Computer Science Hebrew.
Lec 22: Stereo CS4670 / 5670: Computer Vision Kavita Bala.
Announcements Project 3 due Thursday by 11:59pm Demos on Friday; signup on CMS Prelim to be distributed in class Friday, due Wednesday by the beginning.
Color and Brightness Constancy Jim Rehg CS 4495/7495 Computer Vision Lecture 25 & 26 Wed Oct 18, 2002.
Shape from Shading Course web page: vision.cis.udel.edu/cv February 26, 2003  Lecture 6.
Multiple Light Source Optical Flow Multiple Light Source Optical Flow Robert J. Woodham ICCV’90.
Announcements Final Exam Friday, May 16 th 8am Review Session here, Thursday 11am.
Large-Scale Matrix Factorization with Missing Data under Additional Constraints Kaushik Mitra University of Maryland, College Park, MD Sameer Sheoreyy.
Lighting affects appearance. Lightness Digression from boundary detection Vision is about recovery of properties of scenes: lightness is about recovering.
Tal Amir Advanced Topics in Computer Vision May 29 th, 2015 COUPLED MOTION- LIGHTING ANALYSIS.
776 Computer Vision Jan-Michael Frahm Spring 2012.
Understanding the effect of lighting in images Ronen Basri.
EECS 274 Computer Vision Sources, Shadows, and Shading.
776 Computer Vision Jan-Michael Frahm Spring 2012.
MAN-522 Computer Vision Spring
CS4670 / 5670: Computer Vision Kavita Bala Lec 27: Stereo.
Merle Norman Cosmetics, Los Angeles
3D Photography: Epipolar geometry
Epipolar geometry continued
Uncalibrated Geometry & Stratification
Reconstruction.
Announcements Today: evals Monday: project presentations (8 min talks)
Shape from Shading and Texture
Presentation transcript:

1 Resolving the Generalized Bas-Relief Ambiguity by Entropy Minimization Neil G. Alldrin Satya P. Mallick David J. Kriegman University of California, San Diego

2 Image Formation Model ■ Camera centered coordinate system ■ Orthographic camera Image plane Surfac e

3 Image Formation Model ■ Assumptions : Lambertian reflectance Orthographic camera Distant point light source No cast shadows / interreflections ■ Image formation,

4 Photometric Stereo ■ [Silver 1980, Woodham 1981] ■ Goal : Recover surface (normal map) Fixed scene Fixed viewpoint Varying Illumination ■ Solve linear system, ■ Solution : (Yale Face Database B)

5 Uncalibrated Photometric Stereo ■ What if the lighting is unknown? ■ Family of solutions, [Hayakawa 1994] [Epstein, Yuille, & Belhumeur 1996] [Rosenholtz & Koenderink 1996] ■ Factorize with rank 3 SVD approximation, Recovers, up to a 3x3 invertible linear transform

6 The GBR Ambiguity ■ SVD + Integrability, [Belhumeur, Kriegman & Yuille 1997] [Yuille & Snow 1997] ■ encodes the GBR ambiguity, 3 parameters (Belhumeur, Kriegman, and Yuille)

7 The GBR Ambiguity ■ SVD + Integrability, [Belhumeur, Kriegman & Yuille 1997] [Yuille & Snow 1997] ■ Surface height, ■ Albedo & normal, (Belhumeur, Kriegman, and Yuille)

8 The GBR Ambiguity ■ SVD + Integrability, [Belhumeur, Kriegman & Yuille 1997] [Yuille & Snow 1997] ■ Surface height, ■ Albedo & normal, Albedo mapAlbedo map (GBR)

9 The GBR Ambiguity ■ SVD + Integrability, [Belhumeur, Kriegman & Yuille 1997] [Yuille & Snow 1997] ■ Surface height, ■ Albedo & normal, Albedo mapAlbedo map (GBR)

10 Resolving the GBR ■ Need additional constraints, Light source strength [Yuille & Snow 1997] Surface reflectance [Drbohlav & Sara 2002; Georghiades 2003; Tan et al. 2007] Surface geometry [Georghiades et al. 2001] Interreflections [Chandraker et al. 2005] Albedo distribution [Hayakawa 1994] ■ Albedo distribution : Only uniform albedo has been exploited previously.

11 Motivation ■ Consider an object with one albedo ■ Distribution well approximated by a delta function Albed o Mas s Albedo map Surfac e

12 Motivation ■ Under a GBR, the distribution is smeared ■ Smearing dependent on and the distribution of ■ Analytic solution : ■ Linear in Albed o Mas s Albedo map Surfac e

13 Motivation ■ Many objects consist of a small set of dominant colors ■ We say such objects satisfy the k-albedos constraint

14 Motivation ■ Multi-color objects have have multiple peaks ■ A GBR transformation smears the peaks True Albedo Distribution Albedo Distribution Under GBR

15 Motivation ■ Does enforcing “peakiness” resolve the GBR ambiguity? Albedo Distribution Under GBR True Albedo Distribution

16 Entropy Minimization ■ Entropy is a natural measure of the spread of a distribution and conversely its peakiness, is a p.d.f. is the support of ■ Doesn't depend on knowledge of k (# of colors) Advantage over most clustering algorithms ■ Entropy minimization has been successfully applied to other computer vision problems [Finlayson et al. 2004; Palubinskas et al. 1998]

17 Entropy Minimization ■ Let be a random variable representing the albedo distribution obtained by correcting for GBR. ■ Let be the p.d.f. of. ■ Objective : Minimize with respect to where

18 Degenerate Configurations ■ k albedos on k planar surface patches True distribution GBR transformed distribution Entropy is constant

19 Degenerate Configurations ■ Nearby albedo modes. True distribution GBR transformed distribution Entropy is lower under GBR

20 Theoretical Correctness ■ Intuition Each albedo corresponds to a delta function in the distribution A GBR perturbs each mode of the distribution Perturbation has finite support proportional to Entropy can only increase as long as modes don't overlap

21 Theoretical Correctness ■ Theorem : The true GBR parameters correspond to a local minima of our objective when the following hold, Surface contains albedo values Non-degenerate surface normal distribution Albedo is independent of surface normal ■ Sufficient (not necessary) conditions. ■ More details in paper.

22 Approximating the Entropy ■ Entropy is defined w.r.t. a density function ■ In practice only have samples drawn from the distribution ■ Options : Fit a continuous distribution (using Parzen windows, mean-shift algorithm, kernel based estimators, etc.) then compute entropy Approximate entropy from a histogram

23 Approximating the Entropy ■ MLE estimator of entropy, is an m-bin histogram computed from n samples ■ is biased, but variance is low ■ used in our experiments

24 Algorithm Overview ■ Input : M images of N pixels each ■ Algorithm i.Recover normals / albedos up to a GBR using algorithm of Yuille and Snow ii.Perform discrete search over GBR parameters a)Apply GBR b)Compute m-bin histogram of albedo values c)Approximate entropy using iii. Apply GBR corresponding to iv. Integrate to obtain surface

25 Experimental Results ■ Input Images ■ Three datasets

26 Stanford Bunny (Synthetic) 6 input images

27 Yale B Face Database 9 input images

28 Red & Yellow Fish 5 input images

29 Review ■ Novel constraint Albedo distribution should have low entropy Valid for large class of real-world objects ■ New method to resolve the GBR ambiguity By minimizing entropy of albedo distribution ■ Validation Theoretical Experimental

30 Questions? 9 input images

31 Future Work ■ Exploit color information Increases separability in the albedo distribution ■ Exploit geometry Facet vectors ( ) should lie on concentric ellipsoids under a GBR ■ Apply method to other ambiguities Lorentz ambiguity [Basri & Jacobs 2001] KGBR ambiguity [Yuille et al. 2001]

32

33