My Submission: Morph Filters

Today I will be showing you the process that allowed me to create Barack Clooney/George Obama.

After some resizing and scaling, I selected 10 points to define the points as follows:

• pupils
• earlobes
• corners of hairline
• corners of chin
• spot between eyebrows
• nose
• center of chin
• middle of hairline
• four corners of image

After determining the points, I also computed Delaunay triangulation of both sets. I ended up using Obama's triangulation, for no particular reason.

I computed the average shape. Then I converted both source images to the new average shape and averaged the two resultant images to get GeObama. Uncanny!

I wrote a generalized function to morph faces. With this, I was able to create a morph animation of george to obama. Cool :D

I computed the average face of the (smaller, 37-sample) Danish dataset.

(for whatever reason, the the face is upside down. However, the later images that use this average face for computation are fine... so I won't question it lol)

I also augmented the data by adding points at each of the four corners, so that the averaged faces in later parts of the computation will also have morphed backgrounds

Now onto those averaged faces! Here are some fun examples of danes whose faces have been morphed to the dataset average:

As you can see, there is a bit of a humorous effect: the averaged face seems to warp people's heads somewhat.

To be fair, that's probably just a result of a lack of data. The face labeling doesn't include any data on the shape of the head beyond the subjects' jaws. As a result the only shape information applied to the subjects' foreheads and above is heavily warped by our corner points.

I also computed the average face of the danes dataset, and mapped it to george clooney's facial shape.

I caricatured the first individual in the danish dataset.

I don't want to sound mean spirited but he does have a bit of a narrow face compared to the averaged danish face. So I suppose it is not unsurprising that his face becomes even more narrow with this transformation

I morphed george to the danish dataset. Unfortunately due to the lack of forehead data george's forehead looks warped.

For a bells and whistles credit, I also computed the average female face, to create a female danish George Clooney. Compared to the average, this George has a few key differences: his lips are thicker and more deeply colored, and his face is rounder.

He also seems to have a more softer expression, especially because of a slight smile compared to the averaged dane George.

I also computed a smiling obama, using my part 1-3 images and this averaged smiling face: