A Google research team has developed an algorithm to eliminate facial sprains that occur when shooting with wide-angle lenses. And it's not just a distortion-correcting filter. The team presented the result of their work at the annual SIGGRAPH 2019 CG Conference.
Ultra-wide-angle lenses, such as those found on smartphones for “wide-angle selfies, ” provide an impressively wide field of view with one drawback - objects are stretched out at the edges. As a result, the background landscape and faces are distorted.
Applying a simple perspective distortion correction filter "tidies up" the image around the edges of the image, but this requires "bending" the perspective lines of the entire photo. This makes faces look better, but any straight lines in architecture or the skyline in the background appear unnaturally curved.
The team's decision is to use different processing methods for the faces and backgrounds. To do this, the algorithm first determines what is shown in the photo and distinguishes between the faces and the background. A perspective correction map is then created for the entire photo, but only applied to the subject areas, thus keeping the background straight while shifting faces in perspective.
According to the developers, the technology works great and fast, which underlines its "interactive" status.