Face Morphing

Approach

Steps:

1. Define the point correspondences for each face
2. Compute the average points between the two faces' points
3. Triangulate each face using the points, with Delauney Triagulation
4. Warp both faces to the average points using affine transformation
5. Cross dissolve the warped faces

Results

Here are two faces from my friends Tara and Myra. I chose these photos because the backgrounds are really similar, so you can see the effect of the warping and morphing for the next step.

Approach

Create at gif with 45 frames. Each frame is a separate morph, which is similar to the mid-way face. The mid-way face is an example of a morph with warp and dissolve fractions equal to 0.5. We create each frame by varying the warp and dissolve fraction incrmentally, within the range 0 to 1. Each frame increases the warp and dissolve fractino by 1/45. These morphs are combined to form a gif, demonstrating morph sequence.

Results

The 45 morph frames are combined to create a smooth morph sequence.

The return of Ednanxi

In project 3, I produced a hybridized image of my friend Nanxi and Edna Mode. I decided to continue on with the Ednanxi legacy by making a morph sequence.

Approach

I used the data set of Danish faces to compute the mean face.

Steps

1. Grab the points of all faces in the collection
2. Compute the mean of these points to be the set of mean points
3. Warp each of the faces in the collection to the mean points
4. Combine those photos, weighted equally

Results

Here are some examples of some of the Danish faces warped to the mean points.

Now to mix my own face to the average Danish face...

Approach

Steps to create a single caricature:

1. Obtain the population mean
2. Calculate the difference between mean face points and face points
3. Scale that difference by some alpha value
4. Add the scaled difference to the face points
5. Warp original face to the new face points

Results

To make for better caricature results, I computed the average female Danish face.

I computed the caricature at varying alpha values. The results were rather unappealing.

Negative alpha values means that the resulting warp is less Danish, while positive alpha values mean the resulting warp is more Danish.

Approach

The goal of this portion was to change a person's appeared gender with morphing. This was accomplished by morphing the average face of a certain gender with a given face. I decided to combine this with ethnicity to see what kind of interesting results we'd obtain. I also tried this on both male and female faces for fun.

Results

I started off by morphing my face to the average Taiwanese male's face. I suppose the result looks more male, but I think the result actually looks pretty similar to what I looked like in middle school :(

Next I morphed myself with the average American female face.

I decided to morph my friend Wilton's face to the same average faces we looked earlier, just to compare the results when starting from a male face.

Personally, I think that Wilton has never looked better than when he was an American female.