Overview

For this project, we use warping, interpolation and cross-dissolving to produce a "morphing" animation between two face images and also calculate image properties like mean face and implement features transfer.

Part 1 Defining Correspondences

Finding corresponding feature points is crutial in image morphing. The corresponding points are manually selected in the same perticular order. In addition, Delaunay triangulation is used to subdivide the points used for interpolation.

Part 2 Computing the "Mid-way Face"

For the source face and target face, we calculate the midway face using coordinates mean and an affine transformation to do the inverse-warping on pixels.

Part 3 The Morph Sequence

Finally, we are able to generate a series of frame images using interpolation on coordinates and cross-dissolving on pixel values, and simulate image morphing

Part 4 The "Mean face" of a population

Using the Danes libaray, we can apply the same technique use for image morphing.

Then we can calculate the mean face of the dataset.

Using the mean face, we can warp some faces in the dataset to match the average shape.

We can also warp images of ourselves to match the average shape of the dataset, or use the mean face of the dataset to match out face.

Part 5 caricatures: Extrapolating from the mean

Starting from the male mean face, we can extrapolate the difference to produce caricature-like images.

Bells and Whistles

Using the image data from the Danes dataset, we can tranform images of ourselves to differnt gender or ethnicity.