After defining keypoint correspondences between a selfie and a portrait of George Clooney, I computed a Delaunay triangulation of the average geometry and created a function that morphs between our two images, interpolating between both the shape and the appearance. Our mid-way face is shown below.
|
|
|
Using the interpolation function created in the previous part, I created a morph sequence of 45 frames, warping our images to an intermediate geometry and dissolving between them accordingly. The morph sequence is shown below.
I decided to analyze a subpopulation of the Danish dataset consisting of the images of all women in a neutral pose. I then warped each image in the subpopulation to the average shape of the set. Some examples are shown below.
|
|
|
|
|
|
The images created in the last step were then averaged to create a representation of the "average" face of the subpopulation, shown below. As you can see, although each individually warped image looked distorted when viewed in isolation, averaging all of the images works quite well.
|
Finally, I warped my face to the average shape of the subpopulation and warped the average face from the subpopulation to my geometry. The results are shown below. As you can see, warping my face to the average subpopulation shape made my chin rounder and my nose look pinched in.
|
|
|
Extending the previous section, I again warped my face to the average subpopulation geometry, except this time extrapolating from the average geometry by choosing a warping constant that was greater than one. The purpose of this was to attempt to make a caricature out of my face, which can be seen by the severely pinched-in nose and bowl-shaped chin below.
|
|
I attempted to warp my face to a representation of the average European male face as found on Google Images. Below, you can see the results of morphing just my geometry to the European average, warping just my appearance, and warping both.
|
|
|
|
|
|