The same problem that search engines faced in the early days when searching text, we now begin to experience in other media such as images or video. While sites like Flickr or Blinx do a very good job at finding relevant results, they are mainly based in captions, tags or metadata, prone to human interpretation and ambiguity.
This was until i found last week the work of a portuguese working at the UCSD School of Engineering, with a interesting research in the fields of media search. Nuno Vasconcelos, a former member of the research staff at the Compaq Cambridge Research Laborator, and now Professor at USCD, explains his research on Supervised Multi-class Labeling in the video below (5 min) .
He was also kind enough to answer me a few short questions that i’ll transcript here:
Armando Alves: Are the applications of your research mainly scientific ( i.e, medical diagnostics) or are there any plans to release it under commercial license?
Nuno Vasconcelos:Right now, we are still in a research stage, and not thinking about commercial deployment. But, in the long run, it should be possible. I am certainly investigating the possibility of applying this technology to medical imagery, but we are still in the process of learning exactly what we can do.
Armando Alves: Would it apply only to images or do you plan to extend it to video as well ?
Nuno Vasconcelos:Could and will be extended to video. That is one of the areas that we intend to address sometime soon.
Armando Alves: How do you address the semantic relationships between 2 different sets of images?
Nuno Vasconcelos:The idea is to learn vocabularies that are large enough to build generic systems, which do not have to be re-trained across different datasets. There will always be a limit to this, since some domains (e.g. medical imagery) can be very specific. But, at least for regular users (Google style queries), we hope to be able to package the vocabulary with the retrieval system, which you would buy as a piece of software. But we are not there yet.
Portugal does have a slight problem in keeping their brightest minds, but it’s always rewarding to find fellow citizens that are being recognized for their work.
Many thanks for your attention Nuno, and best of luck on your research.