Don’t you hate it when you think you’ve taken the perfect digital picture, but something messes it up? It could be someone walking through, or the protruding gable of house. Andhaven’t you ever wished you could just change the picture and remove the offending object to reveal that pristine view? Researchers at Carnegie Mellon University can help. They’ve developed an algorithm that can remove the unwanted bits of photos and replace it withsomething that fits perfectly. The algorithm scans online photo libraries, such as those on Flickr, to help discover light sources, camera position andcomposition in a photo. It then uses this data to then search for objects, such as landscapes or cars, that match the original photo. The researchers hope to create image libraries everyone can useto doctor their pictures. Developed by James Hays and Alexei Efros, the project was displayed at the Siggraph graphics conference in San Diego. “We searchfor other scenes that share as closely as possible the same semantic scene data,” said Hays. First the algorithm performs a broad analysis of images in the database to eliminate allthose which are not appropriate, retaining the closest 200 for further analysis. From there the useful parts of the best 20 images are cropped and added to the original image to find the best fit.According to tests, only 30% of the altered images could be detected. To generate its clip art, the team has used on the net’s LabelMe image library, which has many objects, such as people, trees and cars, cut out and tagged by its users.
Editors' Recommendations
- The Meta Quest just got an awesome new VR accessory
- The new Microsoft Surface Pro 9 is $330 off at Best Buy
- PCs are about to be flooded with new chip competition
- All the new ways the Vision Pro just got so much better
- AMD’s new Ryzen 9000 is slower than its fastest previous-gen chips
