Parametric Head Creation

Making human heads

Everyone wants to build their own ultimate head model—just once. Either you manage to build your ultimate head model, with perfect topology and UV mapping, or you don't. Either way you probably never want to do it again! If you do succeed, then you want to reuse your model again and again. You want to retarget it to any shape head you need. You want to be able to lower and increase the resolution as you want. You also want to keep all the rigging and animation data you build up on one project and reuse it on every new project. You also hate the idea of using someone else's parametric head. You hate the limitations that the software designer imposed. Worst still, you hate the aesthetic decisions some designer made when choosing the base head models and facial features. You're right. I totally agree.

You should have the best, and it should be your choice. You should be able to reuse all your animation. You should have complete control over the topology, and you should be able to access the best of photo-fitting and hand-modeling features. Fortunately, you can.

There is one mathematical approach to parametric head modeling that can do all of this and a lot more—principal component analysis. Let me explain: if you examine a big enough set of similar data then you can do an analysis to see whether there are groupings (principal components) of features that define the variations in the data. If you do this for scanned head data—you take a large sample of full 3D scans (including color data) of people's heads—you come up with about 140 principal components (parameters) that define a human head, including skin and eye color maps. This is the basis for one remarkable piece of software, FaceGen. This standalone application can encode and generate any human face.

It is based on a large set of full-color 3D scans of people's heads. It has no designer bias—it's based on real data. Because it's parametric, you can: load any topology head mesh you like, and everything still works; do automatic fits to photographs and generate fully parameterized heads ready to animate; blend face parameters or randomize them to generate sets of "relatives"; work with the most powerful and efficient manual modeling tools; and choose race, sex, and age properties with ease because it's based on real population data.

am belaboring the parametric point a bit here, but that's because some things are just right—principal component approach to head analysis and reconstruction is one example. If you want a human head, this really is the best way to go. As you will see, I have imported in the head and UV mapping we created earlier into FaceGen, so I get all the benefits of our own head model and mapping without losing anything FaceGen offers.

am using FaceGen as an example of the best features you want from a parametric head modeling system. Whether you will ever use FaceGen or not, the point is that human heads share a lot of common properties and once you have a good head model, you want to reuse it as much as possible—if someone else has done the work for you, then use it and get on with the interesting aspects of customizing, animating, and bringing your character to life—it's not cheating1

FaceGen is a standalone system that outputs models that you can bnng into any software, and it has a free demo version. FaceGen also has a commercial SDK, so you can integrate its features into your own game like Bethesda SoftworXs did with 'The Elder Scrolls IV Oblivion'

100 ESSENCE The Face

A head above

Parametric head creation

How Create Face Parametric

Set All To Zero

F B ^ tar« with 140 different parameters for both shape and

FaceGen encodes the lace wit. ^ ^ parameters. There is support for texture Here, we see just a h rameters. These parameters have been ooth symmetric and asymmetr.cd m ^ ^ |afge databa$e Qf ^^

derived from principal compor _ ^ ^ ^ regjon an(j Qvera|| ^ 3D head scans. The analysis.^ ^ whjch yQU W(|| have tQ twgak shape it does not mcude t e ^ ^ pafameter space; |t doe$ m of your part,cular head model

Generate | Vtew | Camera . Shapejj Texture | Genetic | Tween | Morph | PhotoFit |

Symmetnc ¡Asymmetric

Brow Rtfge mner - down / up

Brow Ridge Outer - up /down

Cheekbones - tow / high

Cheekbones - shallow / pronounced

Cheekbones - thin / wide

Cheeks - concave / convex

Cheeks - round / gaunt

Chin - forward / backward

Chin - pronounced / recessed

Chin - retracted / jutting

Chin - shallow / deep

Chin - small / large

Chin-wide/thin

Eyes -small/large

Eyes - tilt inward / outward

Set All To Zero

Mirror

Q Step 1: Any topology

FaceGen (FG) has 140 parameters that define the head and its coloring. The parameters are ail based on the average head shown here. Because the head is parameterized, we can use any topology mesh we like. Here are the FG2, FG3, and our head models—the Dase texture is also part of the parameterization. I do most work with our 2K model as it s easy to propagate the changes from this model up or down to other res models back in 3ds Max There really isn't any benefit to loading higher-res models into FaceGen when we can propagate changes to any res model ourselves.

102 ESSENCE The Face

A head above

Parametric head creation

FaceGen has a number of high-level controls that let you vary your mode starting from any face shape. These high-level controls are for sex aae caricature (distance from average human); asymmetry and race Here, we see the sex variation around the average human face from verv temale. through indeterminate sexuality to very male It's easy to create v0ur own face (by hand or fitted to a photo) and then generate its brother or sister—or an exaggerated sexier version.

FaceGen varies parameters from the average human face, but doesn't limit you to just the range found in the human population You can exaggerate any characteristic beyond the normal human range, so creating super-sexy or exaggerated characters is easy and it's predictable Predictability is the key thing. Because of its statistical basis, FaceGen will always give a predictable result based on real human morphology—exactly what we want from a parametric head system.

Head TopologyFacegen Skin

Here, we see the results of taking a photo fit of our model and aging her from 25 (the age FaceGen determined from photo analysis) to 65 in 10 year steps. Note, that both the face shape and texture are affected with age. This will work with any model you start with. You can then export the models and use them as shape targets and texture modifiers to tweak your own aging models—like those Paul Fedor shows in his chapter Aging Monika: Painting In the years' with textures alone.

The data set that FaceGen is based on does not include children. If you want to use FaceGen to create children, then you would make a set of age targets based on the average face and use these as additive morph targets in your own software. This works fairly well, although it is restricted to the one topology model you choose to use, and the results will be based on your own guesses of average young faces—not any real population data.

ESSENCE The Face 103

Jana Kurt Rajce

A head above

Parametric bead cfeabo

O s,ep5 Race

The database FaceGen was built on included people ot African European, South East Asian and East Indian decent Because ot this, you can alter a model based on race Here is Momka with her African, European South East Asian and East Indian virtual cousins Interestingly, when you do an automatic photo fit, FaceGen automatically sets the high-level parameters—so you get an unbiased prediction of

FaceGen PhotoFit Step 2 - Select The Image File(s):

Frontal Image (Required)---

se* age asymmetry, and racial makeup of your subject. It's then very easy to 'weak your 3D model to be a "little sexier", or younger This sort of high-level tweaking is very efficient and invaluable in an entertainment environment where you are often asked to "enhance" the features of an accurate representation of someone.

Left Profile (Optional)

Step 6: Fitting to photos

FaceGen does a pretty good job of creating a model from photographs. This is a fully automated process that requires just two (front and profile) or three photographs. It always works, and the fit is pretty good. If you do need to go in and hand-tweak the fit, it's very fast and easy. You will certainly need to hand-tweak the back of head, ear, and neck shape once you export the model into your favorite 3D software. However, where FaceGen really shines is that the fitted face is fully parameterized. FaceGen uses the photographs as a simulation target and varies its facial parameters until it gets a good fit. It then applies a blended version of your photographs as fine detail texture. Other photo modeling software just produces a mesh and texture that looks like your target, but it doesn't know anything about it. By contrast, the fitted FaceGen head can be applied to any mesh you load into FaceGen. It can be used immediately to generate variations of the original as we did with the race and age, and it will work with all your pre-done rigging and morph targets provided within FaceGen.

104 ESSENCE The Face

Q Step f Genetic variation or me benefits of working in a |

jenerate

J r~ v'1— * w ^^ ______—^ r ' ** • ^ w face you can wander through the parameter space by generating random

J iants on your base face Any variant you select becomes vour new race

variations and

O $teP 8 Parameter modeling

Aitn a parametric mooeiei you change youi moOel by varying parameters- not D> editing a mesh in FaceGen. you can do this m the parameter sliders or simpu by Ctrl-dragging the 3D mooei As you 00 this you can see tne changing parameters atiected b> the tegion you cIick on By restricting the modeling to parameter space you get a degree dt control and efficiency tnat is impdssibie any dthei way ft s truly astounding hovs intuitive ano fast hand-mooeiing can be when it s restricted to the human parameter space

- tn0fatt | V»«" | C»m«ra I Shap« | Twture Ganet.c |îween j Morph | PhotoFrt |

. thi Otnerate button to create random faces similar to current face Ci.ck on CTci t0 choose it for the new starling point Use the "Randomr***' shaer to J I^ust the amount of similarity

Randomness

Generate

Low Cheekbones

Gênerait | View | Camera Shape ; Te*1ure Genetic | Tween | Morph | PhotoFit | Symmetnc I Asymmetnc |

Cheekbones - low/Ngh

Cheekbones - shadow / pronounced

Cheekbones - thin / wtde

Cheeks - concave / convex

Cheeks - round / gvuri

Chin - forwerd /beclcwerd

Chin - pronounced / recessed

Chin - retracted / jutting

Digital Camera and Digital Photography

Digital Camera and Digital Photography

Compared to film cameras, digital cameras are easy to use, fun and extremely versatile. Every day there’s more features being designed. Whether you have the cheapest model or a high end model, digital cameras can do an endless number of things. Let’s look at how to get the most out of your digital camera.

Get My Free Ebook


Post a comment