For this project, I focused on morphing my face into the face of the best kicker in the NFL: Younghoe Koo
|
|
Using correspondance points defined on facial landmarks, I was able to generate a Delaunay triangulation of both faces. My results are shown below:
|
|
To compute the "mid-way face" between mine and Younghoe's faces, I first compute the average shape of both faces. Using the triangulation formed from these mean correspondence points,
I morph both both of our faces into the average geometry. I then average both faces'intensities to produce the midway face. My results are shown below:
|
|
|
To create the morph sequence, I generated 45 frames of intermediate face warps. I made sure to scale the warp and cross-dissolve coefficients linearly, with values lieing between [0, 1]. Here is the result:
|
For this part of the project I used the FEI face dataset. I first computed the average face geometry of all the faces, and then warped each face to the calculated geometry. Here are a few examples:
Once all the faces have been morphed to the mean geometry, I average the intensities of all the resulting images to get the mean face of the population shown below. I also morph my face to the mean face's geometry and the mean face to my face's geometry:
|
|
|
To get a caricature of my face, I subtracted the population's mean facial landmark points from my own to generate a set of vectors that describe my unique features. I then add this set of vectors to my own facial landmark correspondences, to generate a 'caricature' facial geometry.
I then warp my face to this geometry to generate a caricature of myself:
|
I decided to turn myself into a woman! To do this, I first found the average face of Han Chinese women online. I then morph my face into the geometry of the average face of Han Chinese women, and average the intensities of both resultant images. My results are shown below:
|
|
|
|
|