CS 194-26 Project 3: Face Morphing

Defining Correspondences

I manually defined around 35 point correspondences (exact number depending on image) for warping and created a Delaunay triangulation of the points from the average of both sets of points.

Computing the Midway Face

The Morph Sequence

The Mean Face of a Population and Caricatures

I used a subset (first 200 images) of the annotated FEI Face Dataset as a population of face images. I found the mean shape (average control points and its Delauney triangulation) and mean face (average of faces warped to the mean shape). Half of the dataset is consisted of faces with smiling expressions while the other half contains faces with neutral expressions. Taking the average face morphed onto the midway face shape of the entire population has relatively clear eyes and nose features while the mouth is more blurred because of the variation in expressions. The average faces of the subpopulations have more clearer mouth expressions and noticeable differences in the eyes as part of the facial expression.

	Mean shape (entire population)

Morphing into the mean face.

Senua and caricatures with average faces.

The first image is morphing the Senua onto the average face shape. The second is morphing the average face appearance onto Senua's face shape. The last image is more of a caricature: morphing Senua's face and the average smiling face with exaggerated features (using -0.2 for the warp fraction and 0.5 for the dissolve fraction).

Bells and Whistles

Morphing to a new population. The irregularities are due to misalignments and human error from manually inputting key facial points. The face morphings show obvious differences in face shape and eye shape between Senua and the average Asian female. Morphing shape only is morphing Senua onto the shape of the average Asian female, morphing appearance only is morphing the average Asian female's appearance onto Senua's shape, and the full morph is the midway face (combination of morphing shape and appearance).