Project 3

1 Defining Correspondences

Let's try morphing my face and Brad Pitt's face. We'll first by defining the points for key features:

The Delaunay triangulation of the average of these points gives us the following:

2 Computing the "Mid-Way Face"

We notice some relics on the edges. This is because the orientation of Brad's face is a tiny bit off, since he's tilting his head to the side. Let's try this on a couple more images.

If the corresponding facial features are situated similarly for both faces, the morphing does look far better.

3 The Morph sequence

Let's make an animation that gradually morphs two faces into each other!

This looks good.

4 The "Mean face" of a population

Now let's first compute the "mean" face of a set of 100 Brazilian individuals:

Now let's see if we can try to make someone more "average". We'll choose this individual here.

The morph sequence, the midpoint, and the warp:

This looks reasonable. Let's do a couple more.

We can also see what I'd look like if I was a bit more Brazilian by warping the shape! On the right, we have also done the inverse, where the mean Brazilian face was warped to my geometry.

5 Caricatures: Extrapolating from the mean

Now I can "subtract" the population mean from my face to create a caricature. In particular, I chose the mean face of Korean men to fit my demographic.

As you can see, the features that deviate significantly from the mean are emphasized. In particular, the slope on my eyebrows are far more steep. It also grew the width of my nose and eyes. Overall it exaggerates the asymmetry on my face (such as the sizes of my eyes), since the population mean will tend to be very symmetric.

6 Bells and whistles

I'm not a big fan of my original photo! I'll make some changes.

First, let's try to turn myself into a woman. I'll get the mean Korean female face and morph it into mine.

It's actually pretty difficult to give myself more feminine features while retaining my original appearance.

Unfortunately I've already applied some ethnicity-changing transformations in my previous example. So I'll try to change my facial expressions a bit. I would prefer to be smiling in this photo.

Unfortunately, this is the averaged smile of UK Parliament members, so it isn't a population mean. It is predominantly Anglo-Saxon and warps my face in a way that doesn't preserve my original features.

What I can do is use another image of myself to warp my face.

This is a photo of myself doing a joking "wow" expression a few years ago. (After the big game win!)

But now my mouth is just stretched vertically! Time to dissolve more of the new photo.

This is the desired result. Great to see more expression in my photo!