last week... why object recognition is difficult, the template model the feature recognition model,...
TRANSCRIPT
Last week...• why object recognition is difficult,
• the template model
• the feature recognition model,
• word recognition as a case study
Today...
• Recognition of 3-d objects
• Recognition of faces
• Attention
Feature Theory: problems
Relationships between features:
= 1 line1 half-circle
These 3 letters share the same set of features… how are they distinguished by the visual system?
• Each geon has non-accidental features:
– features that you can always see b/c they don’t change in different orientations
– co-terminating lines can be seen to coterminate regardless of where the observer is.
– straight lines always look straight
RBC: Matching process
1 - detect elementary features, edges
2 - find non-accidental properties
3 - determine component geons and their spatial relationships
4 - match to memory
RBC: problems
• Similar objects…
• faces, animals and other natural objects are hard to specify with geons
• RBC is pretty much a bottom-up model of object recognition. Top-down effects have not yet been addressed.
• View-point independent
A possibility:
Maybe there are multiple types of objects recognition, and RBC is good for one of them.
That is, perhaps models like RBC are good for pattern recognition cases where simple features are extracted and structural models are constructed from those features.
Another type of object recognition might involve a more holistic or configural coding of object features. This system would be important for situations where objects were not easily broken down into individual features.
Examples of this kind of object: faces, animals...