CS194-26 Project 4: Face Morphing

Michael Weymouth (cs194-26-adc)

For this project I morphed between faces of different individuals in order to compute video warps, showed the correspondences between different faces in a population, and used those averages to produce caricatures of my own face. Then, for bells & whistles, I morphed my face to that of an average Asian male (from an image found online), to widely varying levels of success. I also experimented with various parameters to produce pieces of abstract art of my face.

Mid-Way Face

For the mid-way face and morph sequence below, I decided to morph between my face and that of Matthew Perry, the actor who portrayed Chandler Bing on the TV show Friends. This character was chosen because I’ve been frequently told that I have a similar personality to him, although we look nothing alike, which makes this morph even more interesting.

From left to right: my picture (image A), Matthew Perry (image B), and the computed mid-way face

Notably, the difference in lighting between the images produces an interesting effect down the center of the mid-way face. In the morph below, however, this effect is not as noticeable over the video.

The Morph Sequence

A GIF of the final morph sequence. Not bad!

Mean Face

For this part of the project, I chose to use the Danes image set explored in class for its completeness and included annotations. First, I computed the average face shape from all of the male and female photos of Type 2 in the data set documentation: “Full frontal face, “happy” expression, diffuse light”. Then I warped all of the images to this shape and averaged the pixels. Then, I warped my face into the average geometry, and then warped the average face into my geometry.

From left to right: An image from the dataset (13-2m.jpg), that same image warped to the Type 2 mean face

From left to right: An image from the dataset (25-2m.jpg), that same image warped to the Type 2 mean face

From left to right: An image from the dataset (14-2f.jpg), that same image warped to the Type 2 mean face

The mean face of Type 2 in the population

My face warped into the average shape computed above

The average face warped into the shape of my face

Caricatures

The difference between the mean face shape computed above and my face can then be extrapolated to produce caricature results of varying quality. I display a few selections, along with their t values.

Caricature with t = -0.1

Caricature with t = -0.2

Caricature with t = +1.5

Caricature with t = +1.7

Caricature with t = +2.0. These results are…interesting.

Bells & Whistles

Using an image I found online of an average of Asian males, I was able to morph my face to change my ethnicity to that of an Asian male. The results offer varying levels of success. I also tried playing around with various parameters outside of their normal ranges and ended up with a few abstract art images, found below.

The average of Asian male faces, found online.

The result of warping just the shape of my face to the above shape.

This didn’t look great…

The result of just cross-dissolving my face and the Asian male face.

Not amazing, but getting better…

Finally, the results of warping both the shape and appearance.

This result is great!

And finally, I present some abstract art of my face, generally produced by varying parameters a bit too far.