Face Morphing

Part 1: Morphing Two Faces

In order to morph two faces together, I labeled keypoints on two images. I then took these associated keypoints and created a triangulation from the mean of these points. The half-way face is then a combination of the two images, morphed to the average triangulation.

Here is an example of the triangulation from the keypoints.

Keypoints on my face.

Results

The results vary very heavily on how well the keypoints were labeled and how many I added to the images. In addition, images that were less aligned and with different hairstyles seemed to look less smooth.

Image 1. My face.
Image 2. My brother.
The halfway face.
Morph video

Image 1. My face.
Image 2. My friend Margo. People say we look alike.
The halfway face.
Morph video

Image 1. My face.
Image 2. The zuck.
The halfway face.
Morph video

The Mean Face

The keypoint morphing algorithm also allows us to compute the mean face of a set of images. I created an average image from the males of the Danes dataset. Here is the result.

The average face smiling.
The mean face.
A person morphed to the average triangulation.
Morphing from the average face triangulation.
Another person morphed to the average triangulation.
My face before morphing.
The average face morphed to my keypoints. This is not good for self-esteem
My face morphed to the average male dane. There are no keypoints above the eyebrows.

Charicatures

Using the average points from the previous part, I now create a charicature of my own face by extrapolating from the mean.

My face morphed to extrapolated average points. Extrapolation is 1.5
My face extrapolated from the female danes.
Average female face.
The female face to my face. Not bad.

Bells and Whistles

Here is a morph of me from death glare to smiling.

Image 1. My face.
Image 2. My passport photo.
The halfway face.
Morph video

Themed Sequence

Here's a morph of my entire family.

My family morph sequence.

Original Photos

Me
Mom
Dad
Sister
Brother