Researchers say a picture is worth 330 words
- 05 December, 2006 08:18
Penn State researchers have figured out a way to help computers describe photographs using 330 English words , a breakthrough that could make it much easier to tag and search for images online.
The Pennsylvania State University has filed a provisional patent application on the invention, which is described in a paper called "Real-Time Computerized Annotation of Pictures." A demonstration of the Automatic Linguistic Indexing of Pictures-Real Time (ALIPR) system can be found at www.alipr.com .
The university says that a picture of two polo players could be deciphered by a computer using ALIPR as having to do with "sport" and "horse" and "polo." The photograph could then be searched on by such tags, which are typically uploaded manually with images.
ALIPR works by analyzing pixel content and comparing the information against a knowledge base containing the pixel content of tens of thousands of images, according to Penn State.
The system still has some maturing to do, however. For 98 percent of the images tested, ALIPR-enabled computers delivered at least 1 correction tag of the top 15 selected words. Fuzzy images and those taken at unfamiliar angles can also cause problems.
The system is fast, completing annotations in roughly 1.4 seconds.
ALIPR builds on earlier imaging work based on computational-intensive spatial modeling rather than models of color and texture distribution.