Table of Contents
Finding Objects in Large Collections of Images
Users want things, not stuff
Standard object recognition won’t work
Need to go beyond traditional methods
Image segmentation and grouping
Factors that lead to grouping
Computational grouping is hard
Blobworld: a new way to retrieve images
Represent images based on regions
Polarity and scale selection
Color features for grouping
Texture features for grouping
Grouping: Expectation-Maximization
Use EM to group image regions
Describe regions by color, texture, shape
Querying: let user see the representation
Use Blobworld to classify images
Prototypical blobs for classification
Learn rules for combining blobs
In Conclusion...
|
Author: ViVE Laboratory
Home Page: http://dli.grainger.uiuc.edu/national
|