Project 5 Writeup

David Dufrense (ddufresn)
March 26, 2010

The goal of this project was to implement a face morphing algorithm. Correspondence points were defined on each image, and a Poisson solver was used to determine the inverse warp of the blended image to the source images. The source images were then warped, and cross dissolved to yield the final morphed image. The control points were defined using the method assignpoints.m to define the initial set of points, and addpoints.m to add correspondence points.

The red points have just been added. The green points were previously extant correspondence points.

Morph Sequence

Mean Face

The face morphing algorithm works well on faces that are similar.

And terrible on faces that are not in the same position.

My means of making caricatures, which is moving correspondence points away from the mean, makes small errors in the placement of these points more noticeable.

I also had the difficulty that a given exagerration factor would not produce a noticeable change in some faces, but produce a high distortion in others.

The apple to orange morph did not turn out very well. This is partially due to the lack of obvious features on the fruit. I put most of my correspondence points on the ouside edge.

The overall shape morphed reasonably, but the stem did not transition correctly during the morph.