In this project we morphed faces into other faces by finding corresponding points for each face and the calculating affine transformations between those points.
For the first part of the project I morphed my face into George Clooney's. Below I have included the original images and point correspondences for each person as well as a the midway face between myself and George Clooney.
This is an animation of the morphing process over 45 frames and back again
For the next part we used a database of faces of Danish people and computed the average face by calculating the shape of the average face and then transforming each person to the average shape before computing average of the pixel values to produce the final result.
First lets look at two examples from the database and what they look like when transformed to the average face shape. We can see that the forehead and hair area looks awkward after transformation in these pictures. this is because the point labeling for these faces was focused on the lower half of the face, so the forehead area does not get transformed well by our process
The average person from the dataset:
I then morphed myself into the average face from the dataset: we can see that the forehead area again does not transform well because of the way the point correspondences were defined for this dataset
Transforming the average face from the dataset into my face works a little bit better but not by much.
In this part we modify the face morphing process slightly. Instead of blending together two faces, I found the difference between my face and added it back to my face, weighted by some constant. What this accomplishes is it emphasizes features of mine that are unlike the average face from the dataset. The result of this process as well as the original image are included below.
Finally, I wanted to see what the result would look like if I morphed my face into that of the average Indian woman's. I found a picture of the average Indian woman online (included below), and calculated the morph for shape only, color only, and including changes in both shape and color. The results for all of these morphs are included below.
Morphing the color only:
Morphing the shape only:
Morphing both color and shape:
We can see that the image morphing is not that great even with both the color and shape morphed because the relative sizes of the original faces in each of their frames was not similar enough