Scene completion algorithm generates semantically accurate images

Sep 15th, 2007 | Posted by Dimchevski in Cool Stuff

scene completion algorithm in action

Researchers at Carnegie Mellon University have developed an algorithm that patches missing parts of images by finding similar image regions in net-based libraries such as Flickr and Picasa.

The algorithm analyses scenes to find out light sources, orientation of objects and the position, pitch and height of the camera. Then it crops the useful parts of 20 best scenes and adds them to the image being edited to offer you a best fit.

What I think is pretty interesting is that it does not mereley morph image areas but rather uses environmental markers to generate a semantically accurate possible match that is then morphed into the image.

I guess one possible area of application is travel business. Those glossy brochures and hotel websites will deceive even more.

I wonder how much this will impact photography? Do you think it'll make it obsolete?
In any case - don't you want to fiddle with this software?

» Digg This | » Add to | » Subscribe to the Journal Feed | » Permanent Link

Leave a Comment

Leave a Comment

Please enter your name

Please enter your e-mail. The e-mail address is invalid.

Please enter your comment.

Saving your comment. Please wait...
Saving comment...

Some HTML allowed: <a href="" title=""> <code> <strong> <em> <strike> <abbr title=""> <acronym title=""> <blockquote cite="">