My Facial Keypoints
Roommate Facial Keypoints
Triangulation
For this part, I manually selected facial keypoints on the portrait of myself and my roommate. I then computed the Delaunay on the average keypoint set.
My Facial Keypoints
Roommate Facial Keypoints
Triangulation
In this part, we use the triangulation of the average point set to warp the source images to the mean image. We do this through an inverse affine transformation. For each triangle, I compute corresponding points in the source image, obtain the pixel value through an interpolation function, and retrieve the color pixel values. To compute a merge the midway faces, we can take a weighted average of the two warped images.
Me Original
Dylan warped to the mean
Roommate Original
Roommate warped to mean
Average of two warped images
In this part, I changed the alpha variable from 0 to 1 which specified the weighted average in which to warp the two images to as well as the weighted average used to combine the image (the dissolve). The larger the alpha, the closer the warped image was to me. The smaller the alpha, the more similar it was to my roommate.
Morph Sequence
In this part, I computed the triangulation of the mean facial keypoint set of a subpopulation (more specifically, I used the Danes dataset and took a subpopulation of males). I then warped all the images of the subpopulation to this and averaged it to get the average face of the subpopulation. I also warped my face to the geometry of the average and vice versa.
Average face of the population
Example dataset faces warped to the average
My face warped to average geometry
Average face warped to my geometry
In this part, I created a caricature of my face by extrapolating from the population mean.
Caracature
In this part, I warped my face to the average female face.
Dylan keypoints labeled
Average female face keypoints labeled
Triangulation on average female face
Dylan warped to average female face
