Project 3

Part 1: Defining Correspondences

I picked an image of my own and downloaded one for Daniel Wu (arguably one of the most handsome guys now on earth). I then cropped the images to contain only the face area and to the same size. After, I handpicked 70 correspondence points for each image and added the 4 corners before generating the Delaunay triangulation at the mean of the correspondence points.

Part 2: Computing the "Mid-way Face"

I looped over the triangles to warp both images into the average shape by bilinear interpolation and inverse affine transformation. I then mix them together to get the mid-way image.

Part 3: The Morph Sequence

I computed the intermediate points, warped both images into the intermediate shape, cross dissolved the warped images, and finally generate each frame of the sequence of morphed images.

Part 4: The "Mean face" of a population

Using the annotated faces from the FEI face database (1a-100a), I computed the average shape, morphed all faces into the average shape, computed the mean face, morphed my face into the average geometry, and morphed the average face into my face's geometry.

Part 5: Caricatures: Extrapolating from the mean

I generated the caricature keypoints by extrapolating from the mean with several different α, then warped my face to the caricature keypoints.

Bells and Whistles: Gender Change

I changed my gender based on the Chinese female average face from The Postnational Monitor. I morphed just the shape, just the appearance, and both.