MIT and Google will teach the smartphone to edit photos in real time

Today nobody will be surprised by the ability of some mobile apps to process photos, to impose on them beautiful color and creative filters, different effects. But what if the smartphone will learn how to get the most beauty out of pictures before it is made? That is the task attempted to solve the engineers and Google. The resulting capable of producing HDR image up to 100 times faster than all existing up to this point the algorithms. This allows you to display smartphone HDR picture from its camera in real time.

In 2015, the MIT alumnus Michael Gharbi developed a cunning method of image processing, which allows to significantly simplify this process. Through the deployment of cloud servers is dramatically reduced bus width for data transfer, and also eliminated the need for a high iron capacity necessary for processing. In fact, any smartphone, equipped with a system that will be capable enough to quickly produce a high quality image from the camera, almost without consuming battery power. But he will still have to send data to the cloud and get servers ready shots. And this, as you know, requires a small, but time.

"the Guys from Google have heard about my invention and have shown an interest. They initiated our meeting and invited me to join efforts in further work on the project. We decided to combine the two approaches to create a completely new data processing system. Instead of having to use cloud servers, it was decided to train the system for everything needed and make it thus completely independent", — shares his thoughts Michael Gharbi.

Engineers used machine learning to train the system based on 5000 images, each of which was accompanied by five retouched variants. The system has developed a certain formula that allows you to edit the colors of pixels in an image. Over time, the algorithm has been perfected to such an extent that he learned to deliver enough high-quality images almost instantly. It is possible to tell, in real time. The system is able to run on mobile devices and does not require high computing power. And the battery it, as well as system Michael Gharbi, uses almost none. The results of the research were demonstrated at the exhibition SIGGRAPH.