I was tasked with computing the face morph between nfeltman (myself) and phoskins. The source images and a 50% blend are shown below:
nfeltman | 50% blend | phoskins |
I found that the key to good results was using many control points. I used 119. A full video of the morph can be found here at 30 fps and here at 5 fps, and each frame can be found as a picture here.
I used a database of 18 images. Since data quality is important, I manually adjusted all of the control points to make the meshes consistent with one another. I also added control points around the border of the image. For example:
All of the source images and control points can be found here.
I first split the class into two populations: men and women. The indeces and average faces of these subpopulations and the full population are shown:
Men | All | Women |
[1 3 4 5 6 8 9 10 11 13 15 16 17] | 1...18 | [2 7 12 14 18] |
I additionally split the men into two catagories: those of eastern Asian descent, and those of European and south Asian descent. The indeces and facial averages for thesee subpopulations are shown:
East Asian Men | Non-East Asian Men |
[1 4 9 15 16] | [3 5 6 8 10 11 13 17] |
Finally I show examples of the average face warped to my geometry, as well as my face warped to the average geometry:
Average Face | Average Geometry |
I present a geometric morph of my face that increase my variation from the average:
Base | Charicature (+70%) | Charicature (+120%) |
I present several morphs of the geometry of my face based on subpopulation averages:
Masculinized | Feminized | Male-Asianized |