In this project, we use Delaunay triangulation to warp images and cross-dissolve them, enabling us to morph faces.
Since this process works best with similar faces, I chose to morph between myself and George Clooney:
I then manually selected corresponding points and generated triangulations over the images:
I then used a change of basis to match the corresponding trianglular slices of both images. After I average their colors, I generated a midway image and a gif showing off the morphing.
We can use a similar strategy over many images to produce a "mean" face. For instance, we can compute the average Danish male's face using 40 different Danish men as samples.
We can then morph some of the samples with the average Danish male to make them more Danish.
Since we have the mean Danish face, we can also morph my face with it to make me more Danish.
We can keep pushing this morph to extrapolate my Danish features and caricaturize my face.
I can use the same techniques to take an average Indian woman's face and morph myself with it. When I morph the geometry of my face, it gets too blown up to seem realistic, but surpirsingly the pure color swap seems more realistic.