Google created a new algorithm for portrait images

The upgrade will help to make the portrait even with one camera on the smartphone.

Google создал новый алгоритм для портретных снимков

The latest smartphones have dual rear camera to capture depth and shooting in portrait style with the background separated from the foreground. Google went on a bold step, and now the new phones Pixel will provide the same effect with one lens. The secret of the new device to install autofocus (PDAF) with the function of a neural network. It works through the principle of Parallax: the camera captures two images from slightly different angles, and calculates the movement for the perception of depth. But sometimes this pattern does not work, because the movement of the user’s hand is not enough to make shots at different angles.

Google создал новый алгоритм для портретных снимков

To solve this problem, Google has created an algorithm that adjusts the depth perceived by the PDAF system. “In particular, we train high-precision neural network created in TensorFlow, which assumes the input pixels are PDAF and learning to predict the depth. This new and improved method of determining the depth of ML is based on the fact that the Portrait mode on Pixel 3, “said Google researcher Rahul Garg.

In order to develop PDAF, the researchers used the device of the five cameras, which helped to determine the correct angle and depth for future cameras. Google has released an album, which compares his old and new methods to show improvement within a few generations of the algorithm for portrait images.