| |
|
Research |
|
Work on this concept began with a class project. The basic idea was to segment images into regions and determine the texture of each of these regions. Using a library of textures as training data a neural network is trained to recognize basic textures. We plan to design ontologies that capture spatial relationships between parts of images. The idea is that once the constituent regions in an image have been identified and labeled with words, the ontology can then be used to assert facts about adjacent regions. Rules could then be used to infer that a particular image made up of certain adjacent regions is in fact an image of some real world entity. Majority of this project is largely undefined and is just a bunch of conjectures. We hope to pursue this in greater depth in the near future. Some key ideas are loosely based on the more generic ideas described in Humanist Computing.
|
|
Literature
and Links |
- Literature
- Links
- Color Structure Code: a powerful segmentation utility developed at the University of Koblenz
- Color Texture Analysis: representation of color textures, also developed at the Image Recognition Lab of the University of Koblenz
|
|
