Learning Texture Descriptors
Start date: Jan 1, 2009,
End date: Dec 31, 2011
We aim to develop novel computational models for describing image textures. By image textures, we mean patterns which arise when many similar structures co-occur. Humans find no difficulties in recognizing thousands of different textures, whether they are natural (grass, sand, sea waves) or artificial (textiles, manufactured surface finishes). The basic goal of this project is to develop novel algorithms for learning texture representations from images. The desiderata for the outcome are good performance (ability to successfully recognize and synthesize textures), scalability (efficient, sub-linear representation and recognition of increasing number of textures) and effectiveness (learning new patterns from small number of examples). As an integral part of the project, we also aim to design benchmarking standards which would enable us, and the computer vision community, to evaluate and compare different texture descriptors.
Get Access to the 1st Network for European Cooperation