The Batman's greatest fear is becoming what he fights.

Computational Photography Project 3: Face Morphing

The main focus of this assignment was to morph my face into that of a classmate's. We'll come to that in a moment. First, consider the morph between these two images: http://www.superherotimes.com/newsarchive/003511.php and http://www.superherotimes.com/newsarchive/003848.php

The first thing to do is define correspondence points on the source and destination images. Then we make a triangle mesh over the average geometry (control points), which depend on the frame. Here is an example of the triangle mesh over my face:

For each triangle, we compute a scalene transformation. Next, we use the set of transformation matrixes to warp each image to the common geometry, as such.

Depending on how far into the morph we are, we cross-fade the colors the approbate amount. Halfway through the Batman-Joker morph, we have this image:

With a sequence of these images, we have the complete morph, as seen above.

Classmate Morph

From me to the (literal) next in line:

This is an optimized (and slightly cropped) version of the final morph. Or it was, until they changed the images that would be used to create the final class "morph all the people cycle through the whole class video". So now there's this new one too.

Mean Face

Step one here is to have a consistent labeling of control points for all faces under consideration. From these points we compute the average facial geometry of the class.

Warping every face to this common geometry and then taking a simple color average of those images yields the mean (average, not angry) face of the group:

It is then trivial to do parlor tricks like warping the mean face to the geometry of a particular person (e.g. myself) or a particular face into the common geometry. Actually, that image would have been computed as an intermediate step to getting the mean face!

What's interesting is what we can tell by looking at that intermediate picture. Notice how my hair line has been pulled up in the mean geometry. That means (ha!) that I have longer bangs than most of my classmantes. My jaw also seems to be less square than average. Ohter observations, those pertaining to the ears, say, or that my glass became skewed, we cannot follow up on becasue of control point imprecision. Still, nifty.

You know what, it's my project! I should be way more important to the mean face than anyone else!

Sweet.