The photo-sharing site Flickr is home to about 200 million images, but most of them go largely unseen and unappreciated. A team of researchers from the University of Turin and Yahoo Labs in Barcelona, Spain, has developed a machine-vision algorithm that can identify and highlight beautiful images, enabling it to spot the hidden gems among Flickr's millions of obscure images.
The team began by crowdsourcing human opinion on the aesthetic quality of 10,000 pictures taken from the Flickr database, a mix of popular and unpopular images in four categories: people, nature, animals, and urban subjects. Each image was rated by at least five humans according to five aesthetic categories. Their beauty ratings were used to train the team's machine-vision system, CrowdBeauty, which used criteria such as contrast, brightness, color patterns, and composition to predict the beauty rating of any given image. CrowdBeauty was then turned loose on a database of 9 million images from Flickr that have fewer than five favorites.
The team then crowdsourced opinion on the images selected by the system and found they were rated almost as favorably as Flickr's most popular images.
From Technology Review
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA