The midway face is computing by averaging the shape of the two images and the color. First, we collect the corresponding points in the two images- nose to nose, mouth to mouth, and eyes to eyes. Since the points must correspond to one another, they must be selected in the same order.
With the corresponding points (2 sets of points from the two images), we can now compute the Delaunay triangluations of the corresponding points' midpoint (The point between the eye of im1 and the eye of im2).
Obama
Putin
Mid Point Image
Putin
Pitt
Mid Point Image
This image did not turn out as well because Rihanna hair covered part of her eyes and face so the midpoint face contains the shadow of her hair
Ann
Rihanna
Mid Point Image
The sequence uses essentially the same logic as the mid-way face but uses 46 different fraction values to compute 46 frames of the "mid way" image (with increasing weights). So in the first frame, the first image should take up the entire image and in the last frame, the second image should take up the entire image. The frames in between have fractional weights of both images
Obama
Putin
Mid Point Image
Putin
Pitt
Mid Point Image
Ann
Rihanna
Mid Point Image
Gathered data from 240 images to create the average face of a population- the image below is the result.
Mean Face
Mean Face with my face shape
My face with the Mean Face shape
Midpoint image of my face and Mean Face
1
2
3
We created Caricatures by bringing one of the images (me) further away from the other image. The following images below corresponds to different weights of how far we pulled the image away. As you can see, the modificaitons and effects increase as we increase the weights.
1.3
1.7
2.5
The images below are change of gender picture from me to an average Chinese girl. The other two images morph only the shape and only the color respectively.
Joseph (me)
Average Chinese Female
Midpoint
Shape of Chinese Female
Colors of Chinese Female