Individual Tasks for 2D Project
Jocelyn make notes Jenny maintain webpage (i.e., add meeting notes, appropriate links to resources, put together a summary of the project from Linda's page about the project) Clifford extend collection of images (i.e., look up collections of pictures online, figure out if they have what we need, and determine cost or just give contact info. to Linda to find out cost) What is object recognition and content-based image retrieval about anyway?
Conceptually...
- feature extraction -- need to be robust against:
- transformations (ex. scale)
- noise
- clutter
- occlusion
- feature selection
- recognition method
Implementation/Programming...
- define data structures for 2D images (read any image, in any format, and put into a data structure
- define a data structure for the features
- translation program from XML to C++
- how can we evaluate our algorithm? (Clifford's task)
How do humans solve this problem?
Humans categorize at two levels:
Does it matter how humans do it for our purposes?
- Perceptually -- car comes up, we instinctively know to move out of the way
- Conceptaully -- use prior knowledge to know that objects that have vastly different appearances are, for example, buildings; consider context
What are the approaches?
Generally, we need to develop algorithms from finite data sets
- many different solutions exist to finite sets
- must regularize problem (use constraints) to make a solution feasible
Two paradigms for recognition
Breaking into parts
- given examples of a class of objects, each object type is broken up into parts
- model should work even when some parts of the object are missing
Semantic Labels
- semantic labels are associated with the pictures
Future Meetings
Meetings with Linda will be held at: 1pm on Mondays 2D meetings will be held at: 2pm on Fridays