This section displays facial keypoints selected from two images: one of myself from freshman year, and one of myself today. I used the given correspondence tool to manually label keypoints. Using Delaunay triangulation, a mesh is created for the face, which is used for the morphing process. You can see the original images, keypoints, and triangulation below:
An affine warp was used to create a midway face between the two images by leveraging the triangulation. Below, you can see the original faces and the midway face:
A sequence of morphed images was generated between the two faces, essentially showing the transformation from one image to the other. Below is a GIF showing the full morph sequence:
The mean face was computed from the Danes dataset by averaging keypoints and warping images to the mean shape. Below are some other Dane faces morphed into the mean shape, and examples of my face morphed into the mean face's geometry:
Various caricatures were generated by exaggerating the difference between my face and the average face. Caricatures of my face are created by amplifying the difference between my face shape and the average face shape, scaled by alpha. However, due to differences in image cropping, even a small exaggeration (low alpha) results in an extreme caricature, which shrinks my face and enlarges the corner triangles. Below are the caricatures at different alpha values:
Using the average female face calculated from the women in the Danes dataset, I morphed my face into a woman by changing its shape, appearance, and both. Below are the results:
This project explored facial keypoint selection, triangulation, affine warping, morphing, computing mean faces, and generating caricatures. These methods allow for powerful image transformations!