Overview
In this project, we morph faces into other faces. However, this involves more than simply fading one image into another because the edges would not line up. By selecting corresponding points on 2 images, you can "warp" these points into each other, giving the illusion of the edge moving with it. Then, depending on which image is more prevalent in a frame, you can linearly interpolate from the source images and take a weighted average of the colors at those points.
Given a dataset of faces, one can also compute an "average" of all faces by warping them into a common shape and then cross-dissolving the colors from these separate images. You can then warp other faces into this average face shape, or, with a slight tweak of parameters, create a caricature of a face.
Defining Correspondences
Usage
python define_correspondences.py [path to image1] [path to image2]
Method
First, using
ginput() I select keypoints on the first image in important places for the morph (e.g. eyes, eyebrows, nose, mouse, edge of the face). Then, in the same way with a consistent ordering, I select corresponding keypoints on the second image. These points will transform to each other in the morph, and are the vertices of the triangulation. Then, I run the Delaunay triangulation algorithm on the averaged points using
delaunay. The result is show below, though it shows the face upside-down.
Delaunay triangulation of the averaged keypoints
|
Computing the "Mid-way Face"
Usage
python define_correspondences.py [path to image1] [path to image2]
Results
Joe
|
Julie
|
Mid-way face
|
The Morph Sequence
Usage
python define_correspondences.py [path to image1] [path to image2]
Results
Morph
|
The "Mean Face" of a population
Usage
python mean_shape.py
python mean_face.py
python face_onto_geometry.py
Results
Delaunay triangulation of the average face shape (upside-down)
|
Person 1
|
Person 1 morphed to the average face shape
|
Person 8
|
Person 8 morphed to the average face shape
|
Person 31
|
Person 31 morphed to the average face shape
|
Average face
|
My face morphed onto the average face shape
|
The average face morphed onto my face shape
|
Caricatures: Extrapolating from the mean
Usage
python face_onto_geometry.py
Results
Caricature of my face with alpha = -0.2
|