We as photographers do our best to produce images with the best quality possible, but despite our best efforts occasionally there will be an important low resolution file that needs to be upscaled. Maybe you find an important picture taken years ago on a low resolution compact camera that suddenly needs to be blown up, or a hard drive crashes leaving photos that haven’t been backed up just a memory – except some low resolution copies in your text message outbox. Whatever the reason, if you’ve ever started with an very low resolution photo and attempted to make it an acceptable resolution photo, chances are you’ve encountered a sad digital rendition. Google is working on a technology to help with this issue that utilizes machine learning.
Rapid and Accurate Image Super-Resolution, or RAISR, has learned to make higher quality enlarged images by analyzing many pairs of images – one low resolution and one high – in order to learn to fill in the blanks and create an enlargement with little data that is usable. It can even combat artifacts and works more quickly than old school methods.
Ever forward-thinking, one idea Google mentions about possible uses for this technology other than the obvious is to minimize data required to load very small graphics onto devices like phones and tablets without reducing the ability to zoom in. The small image would be transmitted using data, and then when the viewer wanted to see a larger version, the device would be able to quickly piece together a detailed, higher resolution photo.
As you can see from the examples, RAISR handles some situations better than others. But it can continue to learn from Google’s vast libraries of images and in time, this could be a really useful tool for those of us stuck with a tiny image that we want to make bigger. For more detailed information, check out Google’s blog post about RAISR.