Overview
The goal of this project is to morph from one face to another. We are able to control the perentage of the morphages. We then collect images of each stages/fraction of the morph and put them together into an animated gif.
Defining Correspondences
The first step is define correspondences by consistently select points at key facial features on both images.
Source (Me), Target
Computing the "Mid-way Face" + Morph Sequence
The next step is to compute the mid-way face of two images. We then averaged these two sets of points and then run it through Delanauy traingulation in order to map corresponding triangles from both images. Finlly we morth both the source and destination image into this average shape to produce a "Mid-way Face." Then, we cross dissolve to get the right colors to produce the resulting mid image. To create a smooth animated transition from one image to another, we have to control the "percentage" of the morph. A 50% will create the "Mid-way Face." We are then able to create a sequence of face morphs by steadily increase the morph percentage from 0% to 100%. In doing so, we are able to create a animation for the morphs of faces.
Source, Mid, Destination, Morph Sequence
The "Mean face" of a population
I used the Danish male, neutral expression faces data set. I averaged out the given points of all image, and ran it through Delanau triangulation in order to create the mid-image shape. I then morph all the images into this mid-image shape. Finaly I averaged out the colors in order to create a mean face of my population data set.
"Mean Face"
Some Examples of morphing into the mean face
Morphing myself to the mean face and vice versa
Caricatures: Extrapolating from the mean
I produced caricature of my face by extrapolating from the population mean you calculated in the last step. To create the caricature points/shape, I used (scalar * (avg_male_faces_pts - points_me) + avg_male_faces_pts). The scalar determines the level of exaggeration.
scalar levels: -0.5, -1, -1.5
Bells and Whistles: Changing Gender
I found the average image for Korean females online and tried to morph my face into it's shape in an attempt to change my gender. I find that it did not warp so well at 100% morph (without changing my colors). There are a few artifacts around the chin and eye. I also tried computing the "mid-way face." I think the 50% morph does a much better job at changing my gender than the 100% morph.
Results
