CG for Vision Research - PowerPoint PPT Presentation

1 / 21
About This Presentation
Title:

CG for Vision Research

Description:

Escher, 3 Spheres (x, y, zT) (x, y, zR) (nxT, nyT, nzT) (nxR, nyR, nzR) ([xLT xRT] ... We can start mixing and matching position, normal, and disparity! ... – PowerPoint PPT presentation

Number of Views:105
Avg rating:3.0/5.0
Slides: 22
Provided by: quocv
Category:

less

Transcript and Presenter's Notes

Title: CG for Vision Research


1
CG for Vision Research
  • Lecture 11
  • Winter Term 05-06
  • Quoc Vuong, Franck Caniard, MPI

2
OpenGL Projections Transformations Graphics
Primitives Textures Lighting Material Properties
Vision Research Viewpoint Effect Structure from X
(Motion, Shading, Stereo) Active
Exploration Navigation
GLUT Window Management Input Devices Event
Loop Timing
Fun/Advanced Modeling Animations Mapping (bumps,
terrains) Physics Simulation Sound (OpenAL)
3
Todays Menu
  • Stereo and shading
  • Dissociating position, normal, and disparity
  • Bump mapping
  • cheap realism
  • Stereo and shading integration

4
Stereo - Shading
5
Escher, 3 Spheres
6
// Normals and z from Teapot // Disparity from
Rabbit! // Right Eye glBegin(GL_TRIANGLES) for
(i0 iltn i) glNormal3f(nxT, nyT, nzT)
glVertex3f(xRR,y,zT) glEnd() // Left
Eye glBegin(GL_TRIANGLES) for (i0 iltn i)
glNormal3f(nxT, nyT, nzT)
glVertex3f(xLR,y,zT) glEnd()
We can start mixing and matching position,
normal, and disparity!
7
Limitations
  • Shading does not provide good disparity signal
    because of smooth gradient
  • Correspondence problem in vision
  • Need textures (features)
  • Position and disparity not completely
    independent disparity distorts projected image
  • ok if disparity is small
  • Humans have high stereoacuity
  • Complexity and differences of 3d models will make
    a big difference!

8
Template Code 11
  • Stereo Shading

9
Bump Mapping
10
What do you see?
11
2d images
  • Most of the time, we think of color
  • red, green, blue channels
  • or grayscale intensity
  • But can encode anything!
  • e.g. intensity height
  • Normals!
  • nx, ny, nz -gt R, G, B

12
(No Transcript)
13
Mark Kilgard, 2000, NVIDIA
14
(No Transcript)
15
Components ofbump mapping
  • Normal map
  • Extension of texture mapping
  • Just an image
  • So can do interpolation, sampling, mipmapping,
    etc.
  • Perturbs normal at each vertex on a per pixel
    basis
  • Tangent space
  • Tangents at local surface
  • Normal

16
Simple idea Perturb normal with normal map
  • Original idea by Jim Blinn
  • Change shape of the surface by changing its
    normal
  • Affects shading!
  • Cheap realism
  • Easier than actually rendering bumpy surface
  • Less memory intensive for higher quality
  • Uses texture mapping techniques to smoothly
    interpolate shading per pixel

Mark Kilgard, 2000, NVIDIA
17
Normals ? Normal maps
  • Normals have 3 components
  • Nx, Ny, Nz
  • Components form unit length vector
  • Nx2 Ny2 Nz2 1
  • Lie within 1 to 1
  • Arbitrarily convert to color space!
  • R (Nx1)/2, G (Ny1)/2, B (Nz1)/2 (get
    into 0-1 range)
  • Any arbitrary RGB image can be treated as a
    normal map
  • Take picture of real bumpy surface, and use that
    image as a normal map

18
Tangent space
  • In texture mapping, image is used to change color
    per pixel of a model within object space
  • In bump mapping, image is used to perturb normal
    to affect shading (and shape) within tangent
    space
  • Space formed by the normal vector, and the 2
    tangent vectors
  • Displacement mapping Actually perturb 3d vertex
    position using similar ideas

19
Micro-textures
  • Psychophysics and vision?
  • Shading gradients provide a weak disparity signal
  • Use bump mapping to add micro-textures
    (high-contrast features) to enhance signal
    strength
  • Does not affect global shading across shape
  • Possibly more controllable than just adding noise
    to the normals (as done in Stereo-Shading
    template code)
  • Efficient method to add realism to stimuli by
    adding micro-textures

20
Template Code 11
  • Bump mapping
  • (modified from www.paulsproject.net)

21
This slide left intentionally blank
Write a Comment
User Comments (0)
About PowerShow.com