Tracking the tools that decentralize the media. tools process ideas resources eventsav

unmediated

 

July 26, 2004

Ramesh Jain on Multimedia Search

Ramesh Jain on Multimedia Search:

We do not have any similar structure defined to consider atoms, molecules, and grammars for pictures. Current image search engines that claim to use image attributes use things like histograms or textures which are neither atomic features nor molecular. They are usually aggregates of atomic features. Just imagine how useful it will be if a document was characterized by saying that it has 5396 a's, 9456 b's, 1294 c's, 529 x's, 1289 y's, and 67 z's. Similar techniques are currently tried by people to search images based on their content. What is needed is to define a "language" to describe pictures.

"This requires knowing those patterns and we don't yet know those patterns. Research in many fields have been addressing these problems and once they have concrete answers, it may be possible to build on top of those. But our spoken languages were not defined like that. They evolved by standardizing certain patterns and then building using those patterns. Should we adopt a similar strategy?


Posted by yatta at 12:19 AM