Face Morphing

Eli Bosworth (eboswort)

My Implemetation

I calculated mophed faces by shifting the selected points Pa and Pb to values determined by warpratio*Pa + (1-warpratio)*Pb. That way the warpratio decided whether the morphed face took it's shape entirely from a, entirely from b or from a mix. I put those shift values into a matrix which I filled in using a Poisson fill. This produced two matrices of shift values, one for each input, that I could use to warp the inputs to the same shape. Then I simply blended the colors together, allowing for an input which decided how much of the color should be taken from each image. I picked 41 points for every face, which I think was a mistake. I did it for the first image not realizing that doing it 19 times was going to take quite a while. I think it was overkill.

Here's the mean face of the class, and a bug that I thought was funny

These are some videos I made that first warp the face, then cross disolve, then do both at once:


I tried to use the mean face to exaggerate the features of faces. My method wasn't very sophisticated. I calcuated the shifts needed to warp a face to the mean face and then I applied a negative multiple of those shifts to the image. In some cases in produced good results that actually seemed to accentuate the subjects features, but the failures are still pretty funny. I also produced some inverse caricatures, where I morphed everything towards the mean face and past it. They are also sort of interesting. See if you can guess which is which

My Caricatures and Inverse Caricatures