Camera tool lets users photo-swap

Digital camera users could soon be able to erase unwanted elements in photos, while importing other images to make their shots more convincing.

Researchers say their new computer editing tool trawls websites to find segments of photos that match the context of the original, so they can fit and blend realistically.

Developed by Carnegie Mellon University, the application, dubbed Scene Completion, draws upon millions of photos from the Flickr website that can potentially replace unwanted aspects. .

After discounting a reported 99.9% of images, it then picks the 200 that bare the closest resemblance to the original photo for further analysis.

The algorithm then searches the 200 to see if they have elements, such as shoreline, mountains or buildings, the right size and tone to fit into the unwanted area in the image.

According to the BBC, which saw a live demo of the tool, the useful parts of the 20 best scenes are then cropped, added to the image being edited so the best fit can be chosen.

James Hays and Alexei Efros from Carnegie Mellon University said that in effect, their tool lets users photo-swap instead of having to Photoshop.

They added that unlike traditional photo editing, users can achieve professional-looking results rapidly and with minimal skills.

In an example posted by the researchers, the tool cut out a roof top that was obscuring the full view of a shoreline, but it can be used to edit out anything unsightly, like lorries or people.

The tool analyses images for light sources, textures, camera position and composition in a photo. As a result, just 30% of the photos altered with it were spotted as not being originals.

The researchers said: “It also looks for image segments that make sense contextually – in other words, it wouldn’t put an elephant in a suburban background or a boat in a desert.”

Another system they are developing is called Photo Clip Art. This uses thousands of labelled images from the LabelMe website as clip art that can be added to photos.

A photo showing a vacant street, for instance, could be populated with images of people, cars or even parking meters, sourced from the LabelMe database.

One of the graduates behind the application said matching an object with the original photo and placing that object within the 3-D landscape of the photo is complex.

“But with our approach,” he said, “and a lot of clip art data, we can hide the complexity from the user and make the process simple and intuitive.”


15th August 2007

Related News

Latest News