1. Processing visual information for Computer Vision
Computer vision introduction
- construct scene descriptions from images
- make useful decisions about real physical objects and scenes based on sensed image
Fundamental issues
- Sensing
- Encoded information
- Representation
- Algorithms

2. Fundamental issues
Sensing – how do sensors obtain images of the world?
how do the images encode properties of the world, such as
material, shape, illumination and spatial relationships?
Encoded information – how do images yield information for
understanding the 3D world, including the geometry,
texture, motion and identity of objects in it?
Representation – what representations should be used for
stored descriptions of objects, their parts, properties and
relationship.
Algorithms – what methods are there to process image information and
construct descriptions of the world and its objects?

3. Some topics Motion analysis from 2D image sequences
Image analysis and segmentation
3D shape from 2D (shape from “x”, e.g., shading, texture, motion,..)
Depth perception from stereo
Stereo model (camera calibration)
3D/2D models and matching

4. Camera Model Camera calibration
- obtain the camera parameters by computational procedure based on
the set of image points whose coordinates are known
(Note: it is not necessary to measure the focal length, offsets, angles
of titles/pans)
- 3D affine transformation
-- translation a certain distance
-- scaling
-- rotation about x-axis (y-axis and z-axis) or arbitrary rotation
(1) Monocular image – single camera
- The shape information is contained in shading, texture, contours,
shadow, motion, etc.

5. Camera Model (1) Monocular image – single camera (cont’d) Y P(x,y,z)
I(x’,y’) Z f X

6. Camera Model (2) Binocular image – two cameras (stereo)
-- missing depth information can be obtained by using stereoscopic
imaging techniques
Disparity: D = x’’ – x’
For eL: (f-z)x’ = (x-d)f
eR: (f-z)x’’ = (x+d)f
So:
z = f – (2df/(x’’-x’)
= f – (2df/D)
= f(1-(2d)/D)
P(x, f-z)
(x’, f) eL
x’=0 d
x=0
(x”, f) d eR f z
x” =0

7. Camera Model
(3) Active camera (movable camera) – Pin-hole model

8. Camera Model
(3) Active camera (movable camera)

9. Camera Model
(3) Active camera (movable camera)

10. 3D information reconstruction
3D object information and representation
Shape from “X”
-- shape from shading
-- shape from texture
-- shape from motion

11. Image Understanding High-level vision Graph isomophism
-- find the similarity of the structure of the graph through matching
Dynamic programming and linear programming
-- Solving combinational optimization problems which involves sequential decision making process
-- Solving optimization problem where cost functions and constraints are strictly linear.

12. Knowledge based computer vision
Prior knowledge of certain object/scene will help computer to
understand what it captured
Similar to human vision system
-- E.g., Template matching design
Relational model design
Image training for computing the basis image

13. Knowledge based computer vision
Image acquisition
& processing
Representation
& description
3D understanding
segmentation
recognition

14. Conclusion
Low-level: spatial-domain methods, transform-domain methods, ….
Intermediate level: region-based, edge-based,…… (segmentation…)
High-level: primitive structure, 3D understanding,
object matching, description/modeling, …..