The “Mean Face” of a Population

We obtained the 37-image subset of the IMM Face Database and computed the average geometry of the faces. We added the four corners of the images to the points to keep areas other than the faces in the outputs. For each of the faces, we morphed the image to the average shape. A few examples are shown below.

We then took the warped faces that share the same average geometry and averaged their colors to obtain an average face. We then tried warping our input image in the previous part into the average shape, as well as warping the average face into the geometry of our input image.

Caricatures: Extrapolating from the Mean

We took the difference between the points of our input picture and the average face and added a weighed fraction (alpha = 0.5) of it to the points of the input picture to produce a caricature shape that exagerates the traits of the input picture, as shown below.

Bells and Whistles

Changing the Gender

We obtained an average face of Caucasian females from this website and manually defined the key points for the face. We then morphed our input picture to the average female face by applying the transformations to just the shape, just the colors, and both.