Accions del document
LARCA seminar, Feb 22nd, Jean-Christophe Janodet
Speaker: Jean-Christophe Janodet, Université Jean Monnet
Time: Feb 22nd, 12:00
Place: Room S208, Omega Building
Title: Searching Patterns in Images using Discrete Representations
Summary: Most techniques for images classification are based on numerical vectors in high dimensional spaces. However, such representations may be difficult to extract in the case of small thumbnail images for instance. In this case, we claim that discrete representations, based on strings, trees or graphs are still challenging.
In this talk, we model images with open plane graphs. We show that isomorphism and subisomorphism problems, known to be intractable in general, can be solved in quadratic time on this kind of graphs.
Moreover, we use our algorithms to retrieve patterns within databases containing thousands of images. Finally, we shall give some hints concerning the development of new distances, based on the largest
common subgraph, or on edit operations, that would allow one to classify images using nearest-neighbour techniques.
Time: Feb 22nd, 12:00
Place: Room S208, Omega Building
Title: Searching Patterns in Images using Discrete Representations
Summary: Most techniques for images classification are based on numerical vectors in high dimensional spaces. However, such representations may be difficult to extract in the case of small thumbnail images for instance. In this case, we claim that discrete representations, based on strings, trees or graphs are still challenging.
In this talk, we model images with open plane graphs. We show that isomorphism and subisomorphism problems, known to be intractable in general, can be solved in quadratic time on this kind of graphs.
Moreover, we use our algorithms to retrieve patterns within databases containing thousands of images. Finally, we shall give some hints concerning the development of new distances, based on the largest
common subgraph, or on edit operations, that would allow one to classify images using nearest-neighbour techniques.
