Wouldn’t it be neat to see what a thousand plus images of a sunset combined into one image would look like? Or how about a thousand images of Santa or even a wedding photo?
A new software developed at UC Berkeley by a team of skilled computer scientists are creating just that. Their new system called AverageExplorer, scours the Internet finding all the images fitting in to your search keyword and combines them into one image. The idea came due to the vast amounts of visual data that is contained on the Internet, with a majority of those images never being seen. These images are often referred to as the dark matter of the Internet.
Through research they discovered that since photography was invented, there have been an estimated 3.5 trillion photos taken, of those photos 10 percent were taken in the last year. Facebook reports that 6 billion photos are uploaded each month while YouTube gets 72 hours of video uploaded every minute. That is a lot of images!
The scientists got their inspiration for this program from artist, Jason Salavon, who finds the top 100 images for a search keyword and manually puts them together into one image. When they started the software generated images, they weren’t as distinguishable as Jason’s images were. After tweaking their software to find common attributes in each photograph, their results looked much like what they searched for. Almost containing a painterly look.
Their system allows the user to identify key elements in the search, so for example a car. You would identify the wheels, windshield, mirrors and rear. By selecting those parts then running the software, the computer average aligns its found images based on those markers. Thus resulting in a more accurate looking image.
There is also an added benefit, other than pretty pictures. Berkeley researchers believe that this new system could teach computer systems how to identify items within an image, thus helping search engines find more accurate results.
To view more images and watch a video on how their software works, click here.