Defining Correspondences

We label a series of points on both images to serve as our controll points for morphing

Next we compute a series of delanuay triangulations to produce a grid of triangles that will be mapped from one image to the next in morphing

"Mid-way Face"

Next we need to implement affine warps between each of the triangles in order to map from one mesh of triangles to the defrmed mesh. This is done my mapping to a canonical triangle then back to a target triangle

To demo this we produce a intermediate blend of two faces

The Morph Sequence

Next we produce a video of this showing a progressive sequence of blends, with inverse billinear sampling to ensure smoothness when producing the result image from both the input images

The Mean Face of a Population

See below the mean faces of a few different "populations" which in this cases are different poses of people's faces

Extrapolating and moving along parallel dimensions

By extrapolating away from the mean or movingalong the difference from neutral face to smiling, we can produce a caracature, smiling version, and frowning version of the same face