CS 180 Project 3

Part 1 Defining Correspondences

To morph one face to another, I need to label the corresponding points on both faces with the same order. I use a tool developed by another student in the previous semester to label them and obtain a correspondences json file.

With the correspondences, I take the mean of points and get the Delaunay Triangulation of the mean of points by using scipy.spatial.Delaunay.

Here are portrait pictures of President Bush and President Obama.

Here are the Delaunay Triangulation of the mean of points from plotted on both pictures

Part 2: Computing the "Mid-Way Face"

Now with the Delaunay Triangulation of the mean face, I can take the following three steps to comput the mid-way face:

• computing the average shape (a.k.a the average of each keypoint location in the two faces), which I have done in part 1
• warping both faces into that shape
• averaging the colors together

In this section, I focus on warping both faces into the shape and averaging the colors. To warp two faces, I need to first warp each triangle from a triangle in the source image to its corresponding triangle in another image using affine transformation.

Here is a pseudocode of how I warp triangles

            for each triangle in the Delaunay Triangulation:
                compute the affine transformation matrix
                create mask for the destination triangle
                transform the coordinates from the destination triangle to the source triangle
                warp the pixels from source image to destination image
                return the warped image limited to the triangle's region
        

After warping triangles from both images, I cross-dissolve (blend) them into one final image. The result shows the morphing between President Bush and President Obama

Part 3: The Morph Sequence

Now I have the mid-way face, I can create a morph sequence by morphing the two faces gradually

Here is the morph sequence between President Bush and President Obama in 30 intermediate steps

Part 4: The "Mean Face" of a Population

I use the techniques of morphing through warping triangles to find the "mean face" of a population. I will work on the Danes dataset with 37 images already labeled. The original labels do not contain the corner points. I add them in for better quality.

First, I compute the average face shape of the whole population. Then, I morph each face to the average shape.

Finally, I morph all the morphed images together to obtain the mean face of the population

I warp my face into the mean shape and then the average face into my face's shape

Part 5. Caricatures: Extrapolating from the mean

Now I will create caricatures by extrapolating from the mean face. I follow the formula of new = myshape + alpha*(myshape-averageshape). Here are some caricatures with varying alphas

Part 6: Bells and Whistles: Change gender

I will change my gender by morphing my face to the average Chinese woman face. I apply my face to the morph function I write for previous sections. I change dissolve_frac and warp_frac to obtain morphing to shape only and morphing to appearance only images