2D Correspondence

1
2D Correspondence

• Part 1
• The 2D correspondence problem
• Representation
• Matching
• Blob analysis
• Part 2
• Template matching
• Hough transformation
• Moments
• (Histogram matching)

2
The correspondence problem
Image
Object model

• Is the object present in the image?
• Where is it?
• Multiple occurencies?
• What are the uncertainties?

3
Different applications
Image
Object model

• Matching
• Quality control
• (Pattern) Recognition
• Stereo
• Correspondence problem
• Tracking
• Correspondence
• Matching

3D
4
Generel solution

• Representation
• Matching

5
Image representation and matching

• Relation between representation and matching
• In general the complexity of the two is invers
  proportional
• Example find the rectangle

Obj. Rep.

• Matching gt course on pattern recognition (6th
  semester)
• Mainly present different representations

6
Abstraction level
Object rep.
Image rep.
7
Pixel methods
Image- window
Image

• Template matching
• Histogram matching
• Filters to enhance data
• Gabor, PCA, Wavelets, FFT, etc.
• Processing
• None/mapping

8
Particle methods

• Connected pixels
• Blob analysis
• Feature vector
• Centre of mass
• Bounding box
• Etc.
• Processing
• Thresholding (BW and color)
• Motion segmentation
• Etc.

Image
Particle
9
Low level Token matching

• Input
• Image window
• Particle
• Geometric primitives
• Edges, lines, corners, circles, ellipses, etc.
• Relations
• Particle features

10
High level Token matching

• Input
• Image window
• Particle
• Contours
• Skeleton
• Dynamic contours
• Snakes
• Etc.

11
Geometric model

• Input
• Image window
• Particle
• High level geometry
• Example Body
• Example Face
• Eyes
• Mouth
• Nose
• Etc.
• Relations

12
Issues to consider

• A good representation
• depends on the application
• reduce the amount of data
• keeps the important characteristics
• is easy and fast to extract
• is easy and fast to match
• for the model is easy to learn