Slashdot Mirror
Facial Morphing Software/Techniques?
scraps asks: " I am assisting an evolutionary psychologist who is researching how symmetry and size of facial features affects how others perceive a person's personality. I am looking for is a way to increase and decrease the symmetry of faces (digital images, not the living flesh), as well as increase and decrease the size of facial features while maintaining a "normal-looking" face. In the days of yore, Gryphon's Morph software seemed to be the weapon of choice, but they seem to have slipped into obscurity. I am using Mac OS X 10.2, and have Photoshop 7.0. Any ideas would be most welcome."
If I understand you correctly, you plan to take a raster image and then "morph" those images to new faces. Kai's Power Goo does this (in a crude and primitive way).
An alternative method (not sure how viable or expensive for you on a Mac) might be to take the image in a 3d animation package (like 3dsmax, XSI) and then create a 3D model based on the displacement. There are tutorials and even more specialized tools that show you how to do this. dvgarage.com has some info about them.
This 3D model should then be much more flexible to model your new face than an image.
The drawbacks of this method is that your model will need to be tweaked to be a more accurate model of a face. After that, it shall have to be textured and shaded. Depending on your skill and tools, it might not be photo-realistic enough for your subjects or whomever shall be seeing these new faces.
It seems to me that the algorithm used by Gryphon's Morph was pretty straightforward. Given a first and last frame of identical dimensions, the number of seconds of animation you wanted, and how many frames per second, and it would generate all the in-between states.
Without any additional control points, this was nothing more than a crossfade between the two sources. Individual control points had a correlation on both the first and last image, so you could map important things like the point of a nose, corners of a smile, etc. With control points in place, the pixels generated for the interim states would be calculated both by what frame in the sequece they were for, and where the pixel was, radially, from the nearest control point. Obviously, the more control points, the better looking the morph.
The next thing it added was the ability to draw vectors between pairs of control points for added smoothing. Draw all around the outline of a face, the eyes, mouth etc, and you'd not only have radial calculations from the control points, but points a distance off the line as well. Not sure exactly how they did that... maybe just right angle to the vector and adjusting influence proportional to proximity.
All that said, it seems the toughest thing would be the input method for defining control points. The calculations based off RGB of individual pixels could probably be done with ImageMagick or any other comprable graphics library...
Not that I have the chops to build an application with GUI, and I'm sure the technique I just described is probably patented by someone, but it doesn't seem like the type of app that should have dropped off the face of the earth...
This sig intentionally left justified.
T. Beier and S. Neely. Feature-based image metamorphosis. Computer Graphics (Proc. of SIGGRAPH), pages 35-42, 1992
Search for PCA analysis and "eigenfaces". Here are some "average" faces and some morphs.s ychology/Staff/pjbh1/facepcai1.htm
http://www.stir.ac.uk/Departments/HumanSciences/P
Yeah, it's for the PC, but it is *so* what you want:
http://www.facegen.com/modeller.htm
Real time 3D tweaking of every possible facial parameter.
What were you expecting?