We map key features between Lauren and George's faces.
We compute a triangulation of Lauren and George's face using the scipy.spatial.Delaunay function to get a "nice" triangulation of the correspondence points.
We compute midpoints between each pair of correspondence points, then use those points to make a new triangulation. For each triangle, we compute an inverse warping from the destination to source triangle. We then warp each triangle accordingly.
Then, we cross dissolve the warped images together.
To create the morph sequence, instead of averaging correspondence points, we weight them (1 - alpha) * im1 + alpha * im2. We cross dissolve at the same rate as well.
This morph sequence was produced using 45 equally spaced weights in [0, 1] combined into a 30fps gif.
To create the average face, we average all the correspondences together to create an average face shape. Then, we morph all of the images to the average face shape, then average their appearances together.
I used the Danish face set to produce the following images: 1. the average male dane face, 2. the average female dane face, 3. my face warped to the average male geometry, 4. the male warped to my geometry, 5. my face warped to the average female geometry, 6. the male warped to my geometry. 7. Select members of the data set warped to the average face.
There is some weird warping in the images because my face is not quite straight on, so when warped to the average faces, my right side becomes stretched out. Similarly, when the average faces are warped to me, their right side becomes pushed in.
Overall, we can see that I have a rounder jaw and shorter face compared to the male average danish face. We can also see that I have a more defined jaw and higher eyebrows than the Danish woman. In both images, my nose becomes more pinched, implying my nose is wider than the danish nose.
To create caricatures, we take a mean face and extrapolate beyond it. I.e. we do (1 - alpha) * original + alpha * avg_face for alpha > 1 for geometic warping.
I used the Danish male neutral face to turn myself into the ultra male Dane. The results look quite weird. Features like a longer face and bigger forehead were exaggerated. Images: 1. the avg male dane, 2. caricature alpha = 1.0, 2. caricature alpha = 1.5, 4. caricature alpha = 2.0
I also used the Danish face set of smiling expressions (male and female) to turn my face into an ultra smiling Dane version of myself. The following images are: 5. the avg smiling Dane 6. Caricature alpha = 1.0, 7. Caricature alpha = 1.5 8. Caricature alpha = 2.0
As you can see, the images do become progressively more comical looking. The mouth looks especially strange because the average image has an open mouthed smile, but my lips are closed.
