We all know that ultra wide-angle lenses are now becoming increasingly popular on smartphone cameras. In fact some smartphones offer the similar lenses with front camera.
The focal length of these varies from 13mm to 17mm equivalent which introduces what is called lens-distortion in the images. Well that’s acceptable when you take wide-angle shots of a scene with natural elements or some architecture or buildings etc. Those shots are even considered more appealing in most of the times. But that becomes problem when you take a shot of a group of people with that wide-angle lens.
Wide-angle lenses will always distort the elements which are near the edges of the frame. So if those elements are faces of the people – close to the edges of the frame, they will be distorted as well. That’s what you usually refer to an unwelcomed effect of stretching, squishing or skewing – left photograph in the featured image above.
Now there is a possible software solution to correct those photos artificially right on the smartphone. Thanks to a group of researchers at Google and MIT who have developed an efficient method to fix those images.
The have published a paper that elaborates the algorithm that fixes the effected photographs by only reshaping the faces and bodies but also adjusting the background and adjoining pixels.
The paper also refers to the existing methods of correcting the distortion which somehow used to fix the faces but also introduce artifacts especially on the background.
So the new method actually runs around the existing methods of and achieving all the good things. While the existing methods usually effect the backgrounds in the image. This new method creates a content-aware warping mesh by detecting the areas of the frame with the faces. It then reshapes the faces without effecting the other elements of the photo.
Researchers told that the algorithm is fast enough to work on on the smartphone “at an interactive rate”. They also mentioned that it gave good results when applied on the photos with a field-of-view (FOV) ranging from 70-degrees to 120-degrees.
Here are example photos fixed with the algorithm and they are incredible. Photos on the left are original, and the photos on the right are after the algorithm was applied.
Checkout more shots (167 images) right besides other methods to see the difference for yourself.
