Project 3: Face Morphing!

Tushar Sharma

Defining Correspondences

I used a total of 36 points for my faces. I thought this was a good number because it's not too large so that it's feasible to look at multiple images but not too small that we're losing information. I made sure to have those points include the corners too for nicer background transitions!

For my faces I used some pictures of George Clooney and Barack Obama, mostly because my mother thinks they're very cute. I took the average of the points and used that to make a triangulation which can be applied to either of the images. Basically, the triangulation can give indices for the triangles and we can apply the original image points to get some clean looking triangles.

Obama Mesh George Mesh

Computing the Mid-Way Face

I really thought about this at a triangle level. We have three sets of triangles, one for each image and one for the average. I warp average triangle to the image triangle, and use that transformation to warp every point in the average triangle to get the color value from the image. For the midway face, I did this process for all triangles with both images and then took the average of those two images. Was honestly stunned when I first saw it!

Mid-Way Face

The Morph Sequence

The concept here is pretty simple. By applying what we have learned in our mid way face, we can just cross-dissolve our points as a function of time to get changing triangles and keep warping to those. As a result, we get this!

Morph Sequence

The Mean Face of a Population

I used the Brazilian dataset and got the average shape by averaging all the points. Without having points at the corners, only the face portions became visible. Here are some faces turned into the average shape.

50 avg 50 80 avg 80

Finally, here is the average face, just an average of all the faces morphed into the average shape. Notice the detail on the eyes and the mouth, and also notice the roundness in shape and cheek. If we take someone with a more defined and angular face shape, I think we would see some interesting results. Here is Obama's face morphed to that average shape and the average face morphed to Obama's face shape.

Avg Face obama to avg avg to obama


I took the difference between the obama points and the average points and then added that back into the obama face shape. What we can expect to see is a further enhancing of that difference, which is that angular effect. I regret making this.


Bells and Whistles

youtube link to class video:

I looked at the average Chinese male face I found online. I morphed my face using only appearance, only shape, and both.

Avg Face obama to avg avg to obama

As a parting note: my partner firmly believes that the longer you are in a relationship with someone, the more your face starts to look like one another's. Perhaps I will use some of the things I have learned in this project to assess this hypothesis.