Interactive Face Averaging (final project)

David Tinapple

15-463: Computational Photography
INSTRUCTOR: Alexei (Alyosha) Efros
Computer Science Department
Carnegie Mellon University

The concept for this project is to create an interactive display that calculates the average image of the face that views it. Using a face tracking system, a display with a camera finds the face of the viewer. The face image is cropped and added to a long-run average as well as a shorter running average, consisting of only a set of recent faces.

The viewer is presented with a screen such as this:

As the viewer moves closer to the camera, the face tracker recognizes the face size as increasing. The system is set to present slightly different images based on how close you stand to the camera. When standing close, the system cross-fades the long-run average with the short-running average to more highly weight the short-run, so that the viewer recognizes himself. The farther the viewer stands, the more the cross fade is weighted to the long-run average image.

Over the course of 3 days, many people visited the installation. The video below illustrates the progression of the average, showing the native camera view, the cropped face, and the building average image. Notice how quickly the total average image settles into a fairly steady image.

The image below represents the total average of all the found faces. An improved version might translate each image slightly to better align with the others, leading to a much sharper image. Also, faces can be filtered to only influence the average when the face is a certain size and found in a certain part of the video image.

Here are a few of my favorite shots from the raw data set. People seem to have fun with this, which is a good sign.

And most surprising is the one person who decides to test the face tracker with a drawing of his own. Indeed it does work. Nice job guy.

thanks,

-David