Content:

   1- Introduction
   2- Pre modeling
   3- Egde Planning & Topology
       3.1- Human Face Topology 
       3.2- Topology
       3.3- Modeling rules
   4- Modeling
       4.1- Polygon Modeling
       4.2- Sub-D Modeling
       4.3- NURB Modeling
       4.4- Zbrush/Mudbox Sculpting Modeling
   5- Common mistakes & tips
       5.1- Eye areas
       5.2- Eye balls
       5.3- Eye lashes & tear
       5.4- Nose & lip
       5.5- Chin
       5.6- Ears
   6- Optimizing
   7- Conclusion

References:

Books & DVD:
- "ESSENCE: TheFace Modeling and Texturing" - Ballistic Publishing 2007
 - "D'artiste:  Character Modeling" - Ballistic Publishing 2005
 - Gnomon Workshop tutorial DVD: " Head Modeling in Maya"

Artists:
- SoaLee
- Max Kor
- Steven Stahlberg
- Maxwagyudi
- Francisco A.Cortina

Websites:
 - www.cgsociety.org
 - www.3dvn.com

1-Introduction:

    Creating believable human face is one of ultimate goals of every 3D artist. The process of creating an average human face is nothing new, however to get the realistic and natural looking we have to make the best out of every step we do: from collecting references, planning before modeling, texturing, rendering...etc. In this essay i will only go into studying the modeling process, i will skip talking about fundamental basic things such as how to use tools in Maya or step by step instruction but pay more attention to upper level things such as : work flow, different method in modeling, topology, common mistakes ...etc all the things that you should put your mind on while modeling to create a realistic head model.

2-Pre modeling:

            Before going into modeling process, i researched at how should we set up references and concept arts as these process have a direct  influence  on our models later. The most important things in this part is the accuracy and quality of reference photographs or concept arts, if we under estimate or make mistake in this process, we might face problem in modeling when we realize that the model fit with the front view image plane but does not fit with the side view, or there is chance that our model look exactly fit in orthographic view but in perspective view it doesn't look alike the person in reference and this will consume a lot of time to make any changes once we have moved into modeling part.

            If we are going to take photograph of a real person for references, It is important to take photograph at right angle or else our model will look wrong and we will have to spend considerable amount of time to adjust it. we should also take into account of our camera 's focal length. the longer our focal length is, the better reference photographs we have. Because the longer our focal length is, the closer our view angle to orthographic view ( like front view, side view  which we base our modeling on). However, in theory we never get a orthographic reference photograph due to distortion of view angle so the models in 3D always look a little bit fatter than the real person, but this can be solved by additional editing in perspective view. The other thing is we should use even lighting because all the detail will be washed out or become hidden with too strong or too low lighting.

 Getting reference images into Maya :

    We cant get the raw photograph straight from camera into Maya image plane for modeling yet, because of the deviation during photographing process we have to check whether these image are exactly fit with each other or not. we should correct our reference  images in Photoshop or any other images processing application.

 
( references images from "ESSENCE: TheFace Modeling and Texturing" - Ballistic Publishing 2007 )

3-Edge planning & Topology:

 
( references images from "ESSENCE: TheFace Modeling and Texturing" - Ballistic Publishing 2007 )

        we should always plan before modeling or else we will get lost in maze of edge loops or have too many unnecessary faces, triangle or pole ( vertex that shares more than 4 edges ).initially we have to sketch out the main shape of the head then base from that we plan our edge structure. Having good topology for our face model is even more important than other object in 3D.

Why ?

        Because topology and edge loops play an important part in building up the head volume, characteristic as well as deciding what the head model can do in animation, moreover unlike other parts of character, the head have much more screen time with many close up shots that involve complex facial expression animation. With a good topology, it 's possible for our head model to have realistic expression, if we have messed up topology, our head model cant do a single facial animation and if it can, the expression will look very un-natural. In addition a nice topology will also make it possible to build the model with less geometry but still look realistic and accurate.

 
( Top: "Europa" by SoaLee
Bottom: "The Night Elf" by Max Kor )

This is an example on how bad topology affects your animation deformer. Due to poor skinning and topology, the polygons at the mouth corner are being sketched as bad as they can. 

With a correct topology we can achieve a much better result with less poly count.

3.1- Human Face topology 

       " So what is a correct head topology? how should we know it is good or bad ? "

     In order to answer this question we should go back and look at human face anatomy. As you can see, the shape of human head as well as winkles and expressions are formed by muscles and shape of skull, the muscles that create the nasolabial fold is :

    But these muscles dont help in determining the shape or the topology of the surface or the nasolabial fold but we should know about them as they are the source of what causing emotion.

        The foam rubber is the skin surface which our model is, and the strings are muscle, and the table is skull. so we should study the movement and shape of the outside skin not the muscle.

        So we decide the topology base on the major wrinkles which is formed by facial muscles. so a good knowledge about muscle and facial anatomy is a bonus but our head model might not look realistic if we try to plan the edge structure on exactly every muscles and wrinkles, it would lead to a dead end. it 's better for us to exaggerate the wrinkles on the reference images and build our topology upon that. 

( reference images by Steven Stahlberg at www.cgsociety.com  )

3.2- Topology:

        This is a basic example of how should we sketch out our topology on reference images, we start with the most identical wrinkles then build up intersected edge flows upon that.

( Reference images from "D'artiste:  Character Modeling" - Ballistic Publishing 2005 )

        Since most of human face are about 99 % similar to each other in anatomy, so the topology would not change dramatically. These are some good and bad examples of face topology I found: 

 
    However, we can see that, in those good topology example, each model has a slightly different topology but they still base on the same way of conceptualize.

" Humans are genetically similar to 99.97 percent or something like that. Yes the lines in the face look different, but we all have the same underlying topology, the potential for those lines. It's like lines in the hands of our palms, all slightly different but basically very similar - for instance you'll never see someone with the lines going in the opposite direction, from the thumb to the pinkie."

Steven Stahlberg - "D'artiste:  Character Modeling" 
Ballistic Publishing 2005

( reference images by Steven Stahlberg at www.cgsociety.com  )

"On the guy, the typical doubled-semi-colon shape that's created by the nasolabial folds as they run into the chin, together with the next wrinkle parallel to it. Almost every single person on the planet shows something quite similar. (Although a few have the nasolabial fold hooking up a bit lower on the chin.) "

Steven Stahlberg at www.cgsociety.com

    It 's also better to know that there is no such looping structure in our human face, as we can see in those reference, breaking our face structure into edge loop is for sufficient and simplification of polygon topology. So that the face's  topology in 3D technical term is approximated and there is no absolute single perfect topology for head, it 's depended on our reference and purpose of modeling.  

3.3- Modeling rules:

The rule for human face modeling is the same as other objects: 

    -As low res as possible,

    - Use quad as much as possible and use triangle when necessary, we should not avoid triangle at all cost because having triangle, despite of the fact that  the model might not deform nicely when animated but it will save we unnecessary edge loop and faces.

    - Avoid poles ( where vertex share more than 4 edges, it 's impossible to avoid pole in our model but we should hide these at those area that are less seen in the model because the areas that have pole are not smoothed properly )  these are some area that usually have pole and they are acceptable.  

    - polygon should always be as chubby or square, no long thin nor diamond shaped ( some of these are acceptable to form wrinkles )

    - the edge direction should not run across form of surface or expression. the worst case is topology running at 45 degrees to the decline curve of the surface.  

( Reference images from "D'artiste:  Character Modeling" - Ballistic Publishing 2005 )

also an important things is :

" the most difficult part is finding the balance between having enough detail to achieve what is needed and building as light a model as possible "  

- Francisco A.Cortina - "D'artiste:  Character Modeling" - Ballistic Publishing 2005.

        The more edge and polygon we have, the more accurate and realistic our model is but the more difficult we can edit it. This apply for both still image and animation. With still image sometime it 's ok but it 's a good practice to keep our model light and low res. If we want to have extra details, the key thing lies in the texture part.: especially normal and displacement map.

4- Modeling:

        Not as strictly as  modeling rules, there are many way of modeling a human face, everyone finds their own a comfortable way to do modeling. These are some method that I have found :

4.1- Polygon Modeling: 

        Polygon is the most popular, basic, essential  and straight forward modeling method. And  it varies in different ways :  

( Reference images from Gnomon Workshop tutorial DVD: " Head Modeling in Maya" )

            - Start from a primitive shape ( cube or sphere...etc) : some people start by modifying primitive basic mesh like cube or sphere. We block out the overall shape of the head first and delete some unnecessary face and edge loop to form a basic topology. after that we start to go into detail by adding more edge and face around area like eye, nose, mouth....etc  

( Reference images from Nnso'n: head modeling tutorial at www.3dvn.com )

            - Start from curve/lines : some people prefer to draw curves that run along the most identifiable  lines of reference images, then edit these lines to form a basic head shape. Then do a extrude command around the lines and start building separated  parts of the face then connect them together when done and carry on editing those gap to form a completed face.  

            - From plane : this method can be consider as the first method i mentioned, we start from a plane and we draw a rough topology on it using split polygon tool or any other drawing edge, cut face tool from plugin. Once we have we basic topology done, we can start pulling the point to form a face that match up with the side view image.

        Polygon modeling let us have direct control over edges, faces and vertices, it 's fast and easy to keep track of modeling process however it require a lot of editing work and sometime it does not work efficiently with surface that need high smoothness and accuracy.

4.2- Sub-D Modeling: 

        SUB-D : this can be also called polygon modeling. Not many people choose this method to work with. At the beginning, it is the same as polygon modeling, once we have our basic head model, we can go into detail by sub-divide one more level to have more vertex for editing. we can also go back and forth during the modeling process to edit vertex at different level. the advantage of this method is the hierarchy workflow it allow we to easily edit the vertex at level 0 if we want to change the overall shape of the head while not being confused by the detail that kept in higher level of Subdivide. the other advantage is the low res version of the model doesn't need to be perfect in topology when we go to higher level, all these N-gone will be all quad and smooth without losing any detail.  

4.3- Nurb Modeling: 

        NURB : nurb is great for modeling automotives, symmetrical objects and accuracy surfaces but it can be used for head modeling as well as other organic objects. the work flow also depend on how skillful we are with NURB, we can make the whole model from start to end with NURB, but it will be a hard work with all the detail of the model, we will have to plan every surface patches for the details we want to add to the model. the common workflow is we usually make the basic head model in NURB then convert it to polygon, and  we can chose ether SUB -D or polygon to go from there.

4.4- Zbrush/Mudbox Sculpting Modeling: 

+ ZBRUSH / MUDBOX : an alternative way to modeling a human head is using 3D sculpting software like Zbrush or Mudbox, we can start with Zphere, or any rough head model, then we start sculpting and adding detail to the model. After satisfies with the detail we have added to the model, we can use the retopology tool in Zbrush or resurface it with new topology in maya to create a lighter model but still keep the similar look of our high res model in Zbrush. 
        The advantage is  we can detail our head model with millions of polygon, also  the process of modeling and topology is separated so  we can concentrate on one thing at a time, they said " modeling first, topology later!". Moreover from the high res model we can generate normal map, cavity map, displacement map...etc these will help we a lot in texturing work later. 
        The disadvantage are : This method take a lot a time, we have to create both super high res model and  low res model. In addition we need to have a good sculpting skill in order to sculpt a good high res model with details. But these are only few disadvantage i can find of this method, overall i think it is really good and being used by many people nowadays.

        Among all the modeling method i mentioned above, i find my self comfortable with the basic polygon modeling which i will use to create the base mesh in Maya and then i will bring it into Zbrush for sculpting and re-topology. Finally i export the lowest level mesh in Zbrush back in to Maya and apply a displacement map & normal map generated in Zbrush to make it look like high res model in Zbrush.

5- Common Mistake & tips:

However, despite of what modeling method we use, there are some common mistakes and areas on the face we should watch out.

5.1- Eye areas 

            Because they often say that eyes are the window of the soul, they are the ultimate important element in 2D painting as well as 3D. Viewers always look at the eyes of character first. it 's the eyes that make the most characteristic out of character's face. So it is never redundant when we spend extra time for the eyes.  

        These small detail seem not to be noticeable  from far distance, but in close up scene they make the eyes look much more realistic and soulful.

5.2- Eye Balls 

Bad example of 3D eyes

Real eye photographs

        Many people underrate this part and they only use one single simple sphere for the eye ball, this make the eye look robotic and dollish. Looking at realistic eyeball, we can see that the eyeball look much deeper and glossier. The key is we have to mimic the real anatomy of the human eyeball. the most used method in modeling realistic eyes nowadays is breaking the eye ball into two separate part:

    - the cornea and the eyeball.

        the cornea is like a sphere but more bent forward at the center, and it's transparency with only catch specular, reflection and even refraction.

    -The eye ball is a sphere but sunken inward at the center to form iris.

This detail modeling can make our model 's eyes believable.  

( Reference images from Nnso'n: head modeling tutorial at www.3dvn.com )
( Reference images from Maxwagyudi : Making of Ingrid Bergman  )

        The water part between the eye lid and eye ball should also be modeled to give more realistic looking of the eyes.

5.3- Eye lashes & tear

        The important thing is the natural shape of eye slashes. the shade is depended on the many elements of the eye slashes.

" an important thing to remember is to get to correct amount of bunching and randomness in both length and placement. The lower lid's lashes tend to bunch in smaller numbers, while the upper lid's lashes bunch in greater numbers. The small membrane on the inner corner of the eyelids, or the 'caruncula lacrimalis', is one of the most important elements of the eyes."  

   - Francisco A.Cortina - "D'artiste:  Character Modeling" - Ballistic Publishing 2005.

        I see that many people use image plane with eye lashes texture on it. But this will make the eye lash look flat, and not good enough in close-up shot.I do some extra study of different method in making eye lashes. some people use Hair & fur in Maya or Hair plugin to make this, but it 's still not really food for close-up.I find out that the best way to do eye lashes is : polygon eye lashes !. Despite of what method we use but the output should be polygon. Polygon eye lashes have a better look with more depth feeling and cast nicer shadows.

        we can use Maya paintFX or third party- plugin to make the eye slashes then convert it to polygon. I find myself comfortable with the method of using Maya paintFX hair to model eye lashes then convert it to polygon, this way i can still render the eyelash properly in polygon form while keeping the construct history of PaintFX, i can modify the randomness, the thickness as well as the number of strands ).  

        Beside, as the number of eye slashes is small ( around roughly 350 strands overall in both side of the face ), converting it to polygon is not a big problem nowadays. If we want to animate without lagging, we can make it invisible since it 's a minor part in animation.

5.4- Nose & lips 

    Through researching, I found out some good topology for those areas :

       The edge around nose area form a round curve starts from top of the nose and go along ala and terminate at area where Philtrum meet nostrils. The mistake we often have is that edge loop doesn't go around the nose but grid-like edge loop instead, it make the nose look square and big.
        In the lips area, we should have at least 2 edge loops go out at the lip's corner or else the model's mouth look like being cut. Also, To emphasize the shape of the lips, we should add an extra edge loop around lips areas because without this edge loop  when we smooth or subdivide one more level, the lib shape will be smoothed out.
         In addition we model the head with close mouth but it 's better to have some extra face loop inside the mouse just in case to hide the border edge when character open his mouth.  

5.5- Chin 

        At the chin area we should terminate edge that run from above with a square polygons area. This topology will make we adjust the shape of the chin easier or just in case when our character have a double chin. we should also have an edge loop start from chin and terminate at ear, this will form our character 's jaw bone more nicely.

5.6- Ears: 

        Ear is a very tricky area in the face, it has spiral structure with a lot of detail in, if we are not careful we might end up spending too many polygon for it. what people often do is: they model the ear separately and then adjust its to meet with the head topology, after all they sew the ear with the head model, we may have some triangle around the ear but since the area behind the ear is hardly seen it 's fine. This ear model can be reused for any other models too.

6- Optimizing:

        It 's essential to keep in mind while modeling that whether our model are suitable to other process or not. We should always review our topology one more time before continue moving to next step to check for unnecessary edge loop, triangles...etc. This will make our work easier later.

- UV mapping :  

            It is necessary  to  keep the topology of the model as symmetrical as possible unless we have to model character with half of his face altered. This will help us a lot in UV and texturing whether we do it  in Maya or other software.

- Animation :
    - Review our topology to check whether our character 's face will have a good deform when doing expression or not.
    - If our character will talk or open his mouth, we will have to hide the open edge of the inside lips area and model the interior of the mouth as well.
    - we also have to put extra edge loop in the upper eye lids area so we will have enough polygon for the eyelid to cover the eye ball when character closes his eyes.
    -Just in case that the character might have frowning or glaring expression, the forehead area will form some wrinkles here. unless if we want to do it by texturing, we can add some extra edge loop here so that we can have enough polygon to make those wrinkles. But this is not really necessary if we dont want such a realistic deform, remove it can help we to have a lighter model.

6-Conclusion:

        Modeling human face is a common topic in 3D graphic. It can be said that face modeling is a simple task, but the most difficult things lie in the harmony of details and shape that make up the unique characteristic and natural of the face. A small change can lead to a vast difference, sometimes it 's difficult to evaluate whether the face look right or wrong, if it looks wrong, which part should be edited ?.A realistic and successful face model will depend much on texture, shader and lighting, however these parts can never be success if they don't have a solid base - the model itself.