In this project, I implement the system of face morphing based on triangulation. In addition, I computed the mean face of a given set of facial images and used the mean face to adjust my original face.

In this example, I generate a sequence of morphing images between two faces.

Download the AVI at here.

Download the images at here

Below, I plotted the mean faces of all the females, all the males and all the students. Observe that the mean face of all the males (Column 2) is very similar to the one of all the students (Column 3). This is because we have very fewer female students in the class.

In this example, I try to feminize my face using all the female faces.

The approache I used can be summarized as the followed equations:

where F0 is the original face. M is the mean face of all the females after warping to my original face. P is a matrix whose columns are the PCA bases. I test three choices for computing P. (1) P equals to I, which is an identiy matrix. (2) P is computed by PCA from all the students' faces. (3) P is computed by PCA from all the females faces. Due to the fact we don't have enough female faces to compute the bases, the result refined by female bases (Column 4) is very bad. In addition, I set a = .7 in the experiments.

Below are the PCA bases and the eigen-values.