By Prangan Tooteja
The goal of this task was to label images with points to create a triangular representation of the face. We label 41 points on each image + the 4 corners of the image to define these feature points. Next, we use Delaunay triangulation to create triangles between these key points to add some sort of structure to the image. Below are some examples of the points used alongside those points projected on the face.
|
|
|
|
The goal of this task was to compute a midway face by morphing two images together. There are two parameters to morph over: the color of the images and the shape of the faces. After labelling the images in the "Defining Correspondences" section, we can compute the average shape of the face by taking the mean of the labelled points. To modify the shape of each image to the mid-way face we calculate the affine transformation from the triangle in the mean image to the triangle in the original image. We can use this transformation to get the brightness values for each channel from the original image and populate into the mean shape. For the second dimension, we can cross-dissolve brightness values by taking the mean of each image. Shown below is the midway image between George Clooney and Ryan C.
|
|
|
The goal of this task was to create a morphing sequence between two images. The idea between this was similar to the previous section where we compute the "Mid-Way" Face between two images but instead of taking the mean of the points and colors we take a weighted mean and gradually increase the warp and dissolve parameters as we transition through frames.
|
The goal of this task was to compute the mean face of the population by taking the average of all faces in a population. The dataset I used was the danish researcher database and here is the resulting image. I slightly modified the functions developed in the previous parts to compute these images.
|
Here are some of the Danish researcher faces morphed into the average Danish researcher's face shape.
|
|
|
Here is my face contorted to the average Danish researcher's shape as well as the average researcher's face contorted to mine. The image looks kind of weird because my original image was slightly more zoomed in than the danish researcher's face.
|
|
|
The goal of this task was to extrapolate a mean to make certain features more prominent. For this, I tried to make myself and my friend more danish by extrapolating the average danish face shape found before to our faces. One of the most noticeable differences is that our eyes are noticeably smaller
|
|
For this section, I used the average Englishman's face from this link The images highlight how important alignment is when projecting faces.
|
|
|
|
|
|
|
|
|
|
|
|
|