At the recent ACM SIGGRAPH 2009 conference, Microsoft Research's Visual Computing Group presented SkyFinder, a search system designed to analyze images of the sky. Microsoft's Jian Sun says that unlike most commercial image-search systems, which rely on text associated with the image, SkyFinder provides good results while also allowing the user to interact intuitively with a search engine.
SkyFinder can be used to search for different types of sky images. For example, a user can enter a request in natural language, such as "a sky covered in black clouds with the horizon at the very bottom," and SkyFinder will return images matching that description. http://www.technologyreview.com/web/23213/
SkyFinder analyzes and categorizes images using a popular method called "bag of words" that involves breaking the image into small patches and analyzing and assigning code words that describe each patch. By analyzing the patterns of the code words, the system is able to classify the image in categories, such as "blue sky" and "sunset," and determine the position of the sun and the horizon.
It also is possible to fine-tune a search using a visual interface. SkyFinder offers a screen that allows users to adjust icons to show the desired position of the sun and horizon. SkyFinder then arranges the images on screen, from blue sky to cloudy sky or from daytime to sunset. An image identified by a user can be used as a guide in a more targeted search.
From Technology Review
View Full Article
Abstracts Copyright © 2009 Information Inc., Bethesda, Maryland, USA