Statistical Color Models (SCM) Kyungnam Kim. Contents Introduction Trivariate Gaussian model Chromaticity models –Fixed planar chromaticity models –Zhu.

Slides:



Advertisements
Similar presentations
Epipolar Geometry.
Advertisements

13- 1 Chapter 13: Color Processing 。 Color: An important descriptor of the world 。 The world is itself colorless 。 Color is caused by the vision system.
Computer Vision Radiometry. Bahadir K. Gunturk2 Radiometry Radiometry is the part of image formation concerned with the relation among the amounts of.
3D reconstruction.
1 Color Kyongil Yoon VISA Color Chapter 6, “Computer Vision: A Modern Approach” The experience of colour Caused by the vision system responding.
Surface normals and principal component analysis (PCA)
1 Graphics CSCI 343, Fall 2013 Lecture 18 Lighting and Shading.
Computer Vision Spring ,-685 Instructor: S. Narasimhan Wean 5403 T-R 3:00pm – 4:20pm Lecture #12.
Pattern Recognition and Machine Learning: Kernel Methods.
Photo-realistic Rendering and Global Illumination in Computer Graphics Spring 2012 Material Representation K. H. Ko School of Mechatronics Gwangju Institute.
Image Processing IB Paper 8 – Part A Ognjen Arandjelović Ognjen Arandjelović
Color spaces CIE - RGB space. HSV - space. CIE - XYZ space.
Computer Graphics - Class 10
University of British Columbia CPSC 314 Computer Graphics Jan-Apr 2008 Tamara Munzner Lighting/Shading II Week.
May 2004SFS1 Shape from shading Surface brightness and Surface Orientation --> Reflectance map READING: Nalwa Chapter 5. BKP Horn, Chapter 10.
Uncalibrated Geometry & Stratification Sastry and Yang
Objectives Learn to shade objects so their images appear three- dimensional Learn to shade objects so their images appear three- dimensional Introduce.
CS292 Computational Vision and Language Visual Features - Colour and Texture.
Basic Principles of Surface Reflectance Lecture #3 Thanks to Shree Nayar, Ravi Ramamoorthi, Pat Hanrahan.
Object recognition under varying illumination. Lighting changes objects appearance.
Information that lets you recognise a region.
Introduction of the intrinsic image. Intrinsic Images The method of Finlayson & Hordley ( 2001 ) Two assumptions 1. the camera ’ s sensors are sufficiently.
1 Numerical geometry of non-rigid shapes Non-Euclidean Embedding Non-Euclidean Embedding Lecture 6 © Alexander & Michael Bronstein tosca.cs.technion.ac.il/book.
Photometric Stereo & Shape from Shading
09/16/02 Dinesh Manocha, COMP258 Surfaces Locally a 2D manifold: i.e. approximating a plane in the ngbd. of each point. A 2-parameter family of points.
CHAPTER 9 Lighting © 2008 Cengage Learning EMEA. LEARNING OBJECTIVES In this chapter you will learn about: – –Light sources – –Point lights – –Spotlights.
X y z Point can be viewed as intersection of surfaces called coordinate surfaces. Coordinate surfaces are selected from 3 different sets.
Lecture II-2: Probability Review
Computer Vision Spring ,-685 Instructor: S. Narasimhan PH A18B T-R 10:30am – 11:50am Lecture #13.
CS 445 / 645: Introductory Computer Graphics
Reflectance Map: Photometric Stereo and Shape from Shading
Lecture 11 Stereo Reconstruction I Lecture 11 Stereo Reconstruction I Mata kuliah: T Computer Vision Tahun: 2010.
Image Formation. Input - Digital Images Intensity Images – encoding of light intensity Range Images – encoding of shape and distance They are both a 2-D.
A fully automated method for segmentation and thickness determination of hip joint cartilage from 3D MR data Authors: Yoshinobu Sato,et al. Source: Proceedings.
Multimodal Interaction Dr. Mike Spann
Camera Geometry and Calibration Thanks to Martial Hebert.
ECE 8443 – Pattern Recognition LECTURE 03: GAUSSIAN CLASSIFIERS Objectives: Normal Distributions Whitening Transformations Linear Discriminants Resources.
Chapter 21 Gauss’s Law. Electric Field Lines Electric field lines (convenient for visualizing electric field patterns) – lines pointing in the direction.
Intelligent Vision Systems ENT 496 Object Shape Identification and Representation Hema C.R. Lecture 7.
6. COLOR IMAGE PROCESSING
Graphics Graphics Korea University Mathematics for Computer Graphics Graphics Laboratory Korea University.
Geometric Camera Models
Course 9 Texture. Definition: Texture is repeating patterns of local variations in image intensity, which is too fine to be distinguished. Texture evokes.
Course 10 Shading. 1. Basic Concepts: Light Source: Radiance: the light energy radiated from a unit area of light source (or surface) in a unit solid.
Resolution Limits for Single-Slits and Circular Apertures  Single source  Two sources.
1 Research Question  Can a vision-based mobile robot  with limited computation and memory,  and rapidly varying camera positions,  operate autonomously.
Edge Detection and Geometric Primitive Extraction Jinxiang Chai.
Photo-realistic Rendering and Global Illumination in Computer Graphics Spring 2012 Material Representation K. H. Ko School of Mechatronics Gwangju Institute.
Artistic Surface Rendering Using Layout Of Text Tatiana Surazhsky Gershon Elber Technion, Israel Institute of Technology.
EECS 274 Computer Vision Affine Structure from Motion.
November 4, THE REFLECTANCE MAP AND SHAPE-FROM-SHADING.
Introduction to Computer Graphics
ECE 638: Principles of Digital Color Imaging Systems Lecture 3: Trichromatic theory of color.
Multiple Light Source Optical Flow Multiple Light Source Optical Flow Robert J. Woodham ICCV’90.
3D Object Representations 2011, Fall. Introduction What is CG?  Imaging : Representing 2D images  Modeling : Representing 3D objects  Rendering : Constructing.
1Ellen L. Walker 3D Vision Why? The world is 3D Not all useful information is readily available in 2D Why so hard? “Inverse problem”: one image = many.
Instructor: Mircea Nicolescu Lecture 5 CS 485 / 685 Computer Vision.
CSC2535: Computation in Neural Networks Lecture 7: Independent Components Analysis Geoffrey Hinton.
11/25/03 3D Model Acquisition by Tracking 2D Wireframes Presenter: Jing Han Shiau M. Brown, T. Drummond and R. Cipolla Department of Engineering University.
Presented by 翁丞世  View Interpolation  Layered Depth Images  Light Fields and Lumigraphs  Environment Mattes  Video-Based.
Computer vision: geometric models Md. Atiqur Rahman Ahad Based on: Computer vision: models, learning and inference. ©2011 Simon J.D. Prince.
1 Resolving the Generalized Bas-Relief Ambiguity by Entropy Minimization Neil G. Alldrin Satya P. Mallick David J. Kriegman University of California, San.
MAN-522 Computer Vision Spring
We propose a method which can be used to reduce high dimensional data sets into simplicial complexes with far fewer points which can capture topological.
LECTURE 10: DISCRIMINANT ANALYSIS
Computer Vision Lecture 4: Color
Video Compass Jana Kosecka and Wei Zhang George Mason University
Bayesian Classification
Mathematics for Computer Graphics
Mathematics for Computer Graphics
Presentation transcript:

Statistical Color Models (SCM) Kyungnam Kim

Contents Introduction Trivariate Gaussian model Chromaticity models –Fixed planar chromaticity models –Zhu and Yuille’s SCM –Oriented planar model –Healey’s model –Bingham model Intensity models Other types of model

Introduction Dichromatic and unichromatic reflection models –Dichromatic: Surface reflected light: Varying with the geometry of the scene, i.e., relative viewing, illumination and surface normal directions. The angular distribution of this geometric factor tends to be strongly peaked around the specular direction. Body reflected light: Arises from subsurface reflection where light enters the body of the material and gets scattered and partially absorbed by subsurface pigment before eventually being reflected back out of the surface. s( ): surface reflectance SPD(spectral power density), i( ): illuminant SPD, m s (g) and m b (g): coefficients, g: a vector defining scene geometry –Unichromatic: Outside the specular peak, the scene radiance (L) is dominated by the body reflected light. This model in fact is often sufficient since specular highlights are localized, sparse (can be considered noise).

Introduction Sensor model: A light sensor samples the SPD of the light, L( ), entering the camera to produce its response. This process is well modeled by a spectral integration. The sensor response x is defined: If the reflection is unichromatic, x is constrained to the positive segment of the line passing through the origin parallel to x b. The direction of this line depends only on the shape of the sensor response functions, surface reflectance and illumination SPDs. Lines radiating from the origin of the sensor response space may be considered to be chromatic equivalence classes. The chromatic properties or chromaticity of a surface under a fixed illumination is thus characterized by the direction of the ray joining the origin to x. The distance from the origin to characterizes the intensity of the data.

Introduction (1) Sensor noise, (2) variations in the SPD of the incident light, and (3) variations in the reflectance SPD will perturb x (intensity and chromaticity). These perturbations give rise, given surface patch, illuminant and sensor, to statistical variations in the color of the corresponding image region.  needs statistical modeling of color.

Trivariate Gaussian model A common statistical model is provided by the multivariate Gaussian distribution. The parameters of the distribution, the mean and covariance, are straightforward to estimate from a sample distribution of data by maximum likelihood estimations. The fixed likelihood boundary has the ellipsoidal shape characteristics of the multivariate Gaussian distribution.

Chromaticity models Assuming unichromatic model of reflection, the intensity and chromaticity of the data are independently distributed. For an approximately single object, we expect the chromaticity distribution to have a single mode. which is peaked about a mean direction. Chromaticity based SCMs generally consist of a Gaussian model of normalized RGB data (chromaticity). These models have conical contours in the RGB color cube with elliptical cross-section.

Fixed Planar Chromaticity Models Represent chromaticity by the point of intersection of the corresponding line through the origin of the RGB cube with a fixed plane, ex., unit plane. Normalized color coordinates are obtained by taking the L 1 norm of the RGB vector: Model chromaticity distribution with 2D Gaussians in the normalized color space (NCS).

Zhu and Yuille’s SCM Ensure a consistently tight model by choosing a plane whose normal is aligned with the mean direction of the distribution (x f ). Its contours in the RGB cube are cylindrical rather than conical. This strategy has the effect of down- weighting low intensity data. Chromatic component of low intensity data is more sensitive to noise (poor resolution in the dark corner of the RGB cube).

Oriented Planar Model Unlike Zhu and Yuille’s model, chromaticity data is obtained by mapping each RGB vector to the point of intersection of the ray from the origin to that RGB point with the plane perpendicular to the mean of the sample distribution. The distribution of points in that plane is modeled by a 2-D Gaussian.

Healey’s Model A more natural and consistent representation for the direction of a ray from the origin is by the L 2 unit vector in that direction (by a point on the unit sphere). Based on L 2 normalization of RGB data. First a trivariate Gaussian model is constructed. The intensity component of the SCM is then made uniform by L 2 normalizing the parameters of the Gaussian model. The mean of the model is found by normalizing the mean vector of the trivariate Gaussian model and the covariance matrix is found by normalizing the trivariate covariance matrix by the squared magnitude of the mean.

Intensity Models Different definitions of the intensity, |x|, that are consistent with the individual models need to be adopted. For the spherical model – Healey model For the Gaussian NCS model based on L 1 norm For the oriented planar, and Zhu and Yuille models Uniform intensity model: highly non-rigid body. no knowledge about the distribution of intensities. Gaussian intensity model: rigid object. The combination of a Gaussian intensity model with any directional chromaticity model.

Other Types of Model Many common color spaces in which the chromaticity and intensity components are explicitly separated. HSV space is a non-linear transformation of the original RGB space. Fitting a Gaussian distribution in the chromatic subspace (H and S) is generally a poor choice since the non-linearity of the transformation can greatly complicate the shape of the chromaticity distribution. Hue is an angular measurement which contains a jump discontinuity at 0 degree (pure red) and is undefined for all shades of grey (R=G=B). Non-parametric SCMs is more accurate when sufficient sample date is available.