The first version of my program took my face and made it smooth. I thought this was pretty cool, and my friend Eileen Lee told me it reminded her of noppera bo monsters (faceless ghosts). However, I realized that I tailored the program very specifically to my own face. It wasn’t smooth enough when I tested it on people with a visible nose bridge, but by trying to remove the nose bridge I ended up erasing my glasses. It was also hard to erase things like cheekbones, since it might end up blurring into things outside of the face and produce the wrong colors. Pretty much, it worked well for me because my face is flat to begin with. :’)
I made a second version using a similar idea, but instead of choosing areas to smooth based on facial landmarks, the user can select which areas need to be “corrected” using colored markings. I was heavily inspired by the Peter Campus performance where he paints green paint on his face and uses chroma key to show another image of his face underneath. I think this version is better for a performance because you can slowly increase the amount of smoothing on your face.
I made this project because I was thinking about having insecurity about your face, and also our weird obsession with “nice” or “smooth” skin. When I see my relatives, they usually greet me by commenting on my face or skin. Things like this often make me wish I could hide my face completely.
All of the blending was achieved with OpenCV seamless clone, and the face tracking in the first version was done with dlib. Special thanks to that school cancellation that gave me 5 more days to work on this!
