A face morph is a seamless transition from one face to another. The first step in creating a face morph is to define pairs of facial correspondences, or landmark points on each face. From these correspondences, I produced a Delaunay triangulation of each image using scipy.spatial.Delaunay.
Each of these triangles corresponds to another triangle in the other face. Taking the vertices of a triangle, I defined an affine transformation of its points to the points in the other image. To compute the midway face (right), I averaged the two faces' geometries and warped each face to this shape. Overlaying the two warps created the midway image.
George warped
Lindsay warped
Georsay
Instead of averaging the two images in shape and color, I assigned each image a weight and produced 45 images, starting with George at 100% and Lindsay at 0%, and gradually increasing George's weight/decreasing Lindsay's weight. Stitching these images together, then reversing the sequence achieves the seamless morph effect.
Using images from the FEI Face Database, I created the mean face of a population with 200 faces. This was done by averaging the facial geometries via triangulation, as well as averaging the color (in this case, black and white intensity). The resulting average face is below.
Here are some example faces from the database, each morphed into the average face shape.
Original
Morphed
Original
Morphed
Original
Morphed
Original
Morphed
I morphed my own face into the average face (shape only). I also morphed the average face into mine.
George
George morphed to average
Average Face
Average Face morphed to George
I made a caricature (right) of my face via extrapolation. I weighted my own face shape by 1.5 and the average face shape -0.5. I then warped my face to this face shape. This enhanced my own unique features, such as narrower face width, larger forehead, and smaller mouth.
We made a face morph video with some of the students in the class! In the video, I morphed from Lindsay to myself, and I also took headshots for many of the students in the chain.
Course project for CS 194-26 Image Manipulation and Computational Photography.
By George Wang. March 2020.
Website template: HTML5 Templates