FACE MORPHING

GAURAV SHAH

POINT CORRESPONDENCE

In order to create a morph between two or more images, the first thing we have to do is label corrosponding feature points present in both images and use those points to create a triangulation. Here, we make use of Delauney triangulation computed on the average of each person's feature points in order to obtain nice triangles. Here we will attempt to morph my face into the face of my girlfriend.

The advantage of using triangles is that we can calculate an affine warp matrix to morph the pixels from one triangle to another. With our average triangulation, we can morph each image using the affine warp to get both images into an average shape, and from there we cross dissolve the colors to get a nice midway image. Hover to see the original image.

In addition to a 50/50 morph, we can vary the weight of the feature warp and the cross-dissolving to create a smooth transition between the face of one individual to another. We can then combine these images to form a nice morphing gif.

Here is the same process, this time for Tony Stark and Steve Rogers. Hover to see the images reshaped into the midway image shape.


MEAN FACES

We can use this idea of morphing images to create a morphed average image of a large population. Using a dataset of 30 Danish male computer scientists, we use our warping capabilities to get to the average of all the Danish male computer scientists. First, let us take a look at a small subsample of the faces in our population.




As before, we now compute the average of the feature points and then use it to compute an average triangular representation. From there, we morph each image into an average shape so that we can then blend them. Here are the above sample images, this time warped into the average shape of the whole population.




We then cross-dissolve all of these reshaped faces to get our mean face. Here is a look at the Triangles used, as well as the end result of the averaging. Notice how the final face is more attractive than any of the individuals, as things such as blemishes and marks are gone, the skin is much smoother, and symmetry is greatly increased.




I can also make it so that my face is reshaped to be that of the mean Danish face, and so that the mean Danish face is reshaped to be in the shape of my face. Let's take a look. Hover to see the altered face.




CARICATURES

I can go further in my endeavour to create ugly pictures by using the mean to create caricatures of myself. By subtracting the mean from the photo of myself, I can obtain which features cause me to deviate from the mean, and then weigh those more heavily/lightly to create an exaggerated version of myself. On the left is a caricature that emphasizes my features, and on the right is caricature that emphasizes the Danish features.




We can see that from these caricatures, the features that differentiate me from the Danish mean is the eye size, forehead size, position of the mouth, and ear size. Definitely an interesting thought experiment and not at all something that is going to leave long lasting damage on my self confidence.


GRAND FINALE

If you have seen any Marvel films in the last 10 years or so, you already know of their iconic opening scene. But as with all things, there was still room to improve. I took it upon myself to help them create a new opening, with the faces of all the main protagonists. Pictured in the movie opening below is Captain America, Iron Man, Spider-Man, Thor, Black Panther, Dr. Strange, Nick Fury, Hawkeye, Black Widow, Bruce Banner, Hulk, Gamora, Nebula, Starlord, Draxx, Vision, and Falcon. If Marvel is reading this, please reach out so we can talk compensation.