One of the challenges for image-search tools is making sure the photos are accompanied by the right metadata or tags, making them easier to find in an online database. A newly unveiled Microsoft Research project uses an algorithm that analyzes images of the sky and automatically extracts key attributes -- cloudy sky, blue sky, sunset, horizon, etc. -- to help people search for specific types of photos.
The project, dubbed SkyFinder, is one of several Microsoft is presenting this week at SIGGRAPH 2009, the annual conference of the Association for Computing Machinery's Special Interest Group on Graphics and Interactive Techniques.
SkyFinder is a research project, not a Microsoft product, so the online interface isn't publicly available. But the company's research can often end up making its way into Microsoft products. And given Microsoft's need to innovate in Internet search, we wouldn't be surprised to see this type of approach surface sometime soon in Bing or some other Microsoft offering.
In the meantime, Microsoft has released a video demo and a paper explaining the technology.
The research was conducted using hundreds of thousands of images of the sky, downloaded from Flickr and user groups.
"In an offline indexing process, a set of semantic sky attributes (e.g., category, layout, richness, horizon, sun position) are automatically extracted from each image," the researchers explain in the paper. "Then in an online search, the user can interactively search sky images based on any combination of preferred sky attributes."
Examples of possible search attributes include "a sky covered with black clouds, the horizon at the very bottom” (which returns the image set 'b' above), or “a clear blue sky with a flying object” (which returns image set 'f' above).
Researchers on the project were Litian Tao of Beihang University, Lu Yuan of the Hong Kong University of Science and Technology, and Jian Sun of Microsoft Research Asia.
READ MORE and COMMENT
No comments:
Post a Comment