I used a ginput script to select points from the images, basing my selection on how Professor Efros labelled his own face! Here are my original images:
Here are my triangulated faces:
I then used the triangulation to calculate the midway face. For each triangle, this involved using a triangular mask to process all the coordinates within a single triangle, performing an inverse affine computation (involving the inverse of A, which we solve for through the equation mat1 @ A = mat2), and interpolating to find the colors of missing points (I used the RectBivariateSpline function). After this, I cross-dissolved. Here is the mid-way face (half way between image 1 and image 2):
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
For morphing, I continually created mid-way faces at specified intervals using different warp/dissolving fractions (I set them equal to each other) and linked them together in a GIF. Sampling at 41 different numbers gave a pretty smooth morph between the faces.
I used all 30 guys in the (pre-released?) Danes image set and parsed for their predefined correspondence points. I computed the mean correspondence points, and warped each Dane face to that mean shape. I outputted every 5 faces the result of this warping:
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
I then combined all the warped faces into a mean face by simply averaging the iamges. Here is the average Dane face:
After this, I used my face and manually labelled the correspondence points in the order of the correspondence points in the Dane image set. Here is the original image:
This is my face warped to the mean Dane face (Candace the Great Dane...):
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
The is the mean Dane face warped to my face (Dane the Great Candace...):
I created caricatures by just exaggerating the differences between my face and the mean Dane face and warping to them. Playing around with these values gave interesting results to say the least:
factor = 0.5
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
factor = 1.5
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
factor = -0.5
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
factor = -1.5
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
I found a picture of a white, male baby with a different expression than me, manually labelled his perfect face, and morphed us together to show my true self. This is the original picture of this astoundingly adorable infant:
A cute mid-way picture of us to show my potential as an infant:
WARNING:matplotlib.image:Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
This is morphing just the appearance:
This is morphing just the shape:
This is morphing both, and it's kinda cute:
