I want to do a little project on facial animation.I've found several papers on the net covering the theory,and i was wondering if anyone knows any source(site,book) covering implementation too(preferably in Direct3D).I've seen some examples that use interpolation between different poses,though i was interested in having control points on the face and applying some deformation algorithm.

thnks for the reply.very helpful

cluster based methods (or basically skinning) is generally more useful for games imho mainly because you can use the same skeletal rig to drive the fcial animations on different models (assuming they are all rigged to a single skeleton). The problem with blend shape aproaches is that they tend to require you to remodel every single facial pose for every single model. The other advantage of clusters is that they do not require any additional work over your standard skinned meshes.

The problem with the skeletal model is once you start wanting fine details, you end up with an absurd number of bones, each influencing just a few vertices, and an absurdly complex animation rig. It's also really difficult for artists to handle, when all they want is to move THIS vertex OVER HERE - they have to fiddle around with multiple bones without the face exploding in the process.

The solution is to do both skeletal and morph-based stuff at the same time (you do vertex morphs in the model's rest-pose space, then you apply skeletal animation). The idea is that you use a skeletal system to model the "bones" of the face - the jaw (obviously) and the major moving parts such as lips, tongue, eyebrows, voicebox, major cheek movement. Then you use vertex-morph-based techniques to control the fine details - the eye crinkles, the exact shape of the muscles around the mouth, etc. The nice thing about this is that once the major motions have been modelled by the skeletal animation, the vertex-morph stuff just has to apply "deltas" from it. These are much smaller movements, and they compress far better than putting the large movements directly into the vertex morphs.

So you get the benefits of both. From the skeletal system, you get the large movements in a form that is fairly simple to retarget to different shapes of face, you get pretty sensible blends even between two very different expressions, and you get good compression of those large movements. You also get a simple cheap low-LOD version you can use for more distant models.

But for the fine detail, the deltas are encoded per-vertex, and the artists still get their exact control. Because the deltas are small, they compress really well. And because they're always relative to the major movements (done by the skeleton), you don't get strange lerping results between different expressions, even when they're quite different. All blending of vertex deformation is done in the rest pose of the model, before skeletal animation, so it "makes sense".

You can also use this trick with clothing. The skeletal stuff does the person's skeleton, and then the vertex morphing just controls how the cloth hangs and wrinkles. The other nice thing is you can do standard IK on the skeleton, and the vertex morphing still deltas off that, even though it has no idea the IK is happening. The cloth might hang in a way that isn't quite perfect, but it's still just "cloth wrinkling", it's not lerping through the arms or anything gross like that.

<mini-pimp>Granny 3D has both these built in, including the delta stuff.</mini-pimp>