When you take a photo of the clouds from behind a plane window or from the nice view of your hotel bedroom, don't you just hate it when the picture inevitably ends up with your reflection on it?
A new algorithm developed by Google in partnership with researchers at the Massachusetts Institute of Technology (MIT) can let people erase foreground obstructions in photos and let people end up with the clear, reflection-free photo they wanted to take.
The technology works by letting users take a short video panning along the object they want to take a picture of. The camera has to be moving slightly so it can take multiple frames from several different angles. From this, the algorithm extracts around five to seven of these frames to determine the obstruction and takes pixels from the other images to put in place of the removed reflection to produce a complete image that is free from obstructions.
The algorithm is not just limited to erasing reflections. As the video below demonstrating how it works suggests, the technology is also capable of removing other foreground obstructions, such as a chain-link fence or raindrops on a glass window.
The video shows the algorithm is not 100 percent perfect, since many of the images produce still show a bit of ghosting, where some semblance of the reflection can still be seen. However, the algorithm is far better than other attempts at removing obstructions in photos.
For instance, Samsung has developed its own camera algorithm that is available in the Galaxy S4. Dubbed Eraser Mode, Samsung's technology basically works the same as this one, but it cannot scale up far enough to remove obstructions that take up a good portion of the photo, such as reflections and cyclone fences.
As such, Google and MIT's algorithm still appears to be in the early stages of development, and they likely have to tweak their technology before it appears in cameras and smartphones, that is, if Google decides to bring this to the masses.
The researchers will present their algorithm at the Siggraph 2015 conference. For those interested in checking out the details of the research, the researchers have posted a copy of their paper online.
