Creating 3D Human Character Mesh Prototypes from a Single

Front-view Sketch

Shaikah Bakerman, Rufino R. Ansara and Chris Joslin

School of Information Technology, Carleton University, 1125 Colonel by, Ottawa, Canada

Keywords: Sketch-based Modeling, 3D Geometry, Topology.

Abstract: 3D character modeling, a vital part in film and video game production, is a process that starts by blocking

out the basic geometry of a character, which is then transformed into a detailed mesh. This process can be a

long and daunting task to novice modelers. To facilitate this process, we developed a sketch-based modeling

system that constructs the basic 3D geometry of a human character based on a single front-view sketch and

minimal user interaction. The objective of this system is to help novice 3D artists by automating the initial

phase of the modeling process with an intuitive user interface, and to construct a mesh that conforms to

common modeling techniques. We conducted a user study to evaluate the system’s user interface and

produced mesh. Results show that our interface is intuitive, easy to use and produces accurate meshes that

can facilitate the modeling process. Moreover, the processing time of our system is faster than the average

time it takes artists to manually construct a similar mesh.

1 INTRODUCTION

Three-dimensional (3D) modeling of human

characters is a vital part of the entertainment

industry. Building a 3D character commonly starts

by blocking out the basic shape of the character and

ends by reaching the level of detail desired (Dargie,

2007). Depending on the required level of detail,

character modeling can be complex as it is

constrained by professional guidelines that 3D

modelers are expected to follow to ensure suitable

results (Patnode, 2012).

To organize and facilitate this complex

process, the common rule of building user

interactive 3D modeling tools is user-centric design

(Mao, Quin and Wright, 2006), where producing an

intuitive and easy-to-use user interface (UI) is the

main goal. However, despite the popularity and

increasing use of 3D software, such as Autodesk

Maya, Autodesk 3DS Max and Blender, both experts

and non-experts consider these tools complex and

difficult, and demand better interfaces (Cook, Agah,

2009). Moreover, per Alhashim (2015), novice 3D

modelers find that using existing shapes to model a

complex mesh is easier and less tedious than starting

the process from scratch. This has led to the

emergence of sketch-based modeling (SBM)

techniques that strive to find efficient and accurate

approaches to automate parts, if not all, of the

modeling process.

1.1 Sketch-based Modeling

SBM is the process of automatically creating a 3D

model based on one or more sketches of an object.

One SBM method, called interactive sketching, is

used to dynamically construct and modify a 3D

model using the system’s UI (Andre and Saito,

2011). Another method, consistent with our

proposed methodology, involves providing images

to the system for processing (Entem et al, 2015).

1.2 Research Contributions

Our main contribution is an SBM system that

constructs the 3D geometry of a human character

with a single image. Our goal is to facilitate the

character creation process. To do this, we built a

script compatible with Autodesk Maya, a common

3D modeling application. Our algorithm constructs a

3D model that conforms to the geometrical

constraints of common practices, bringing the SBM

concept closer to the standards of an industry that

248

Bakerman, S., Ansara, R. and Joslin, C.

Creating 3D Human Character Mesh Prototypes from a Single Front-view Sketch.

DOI: 10.5220/0006623702480255

In Proceedings of the 13th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2018) - Volume 1: GRAPP, pages

248-255

ISBN: 978-989-758-287-5

has yet to give its full support to SBM systems

(Olsen et al., 2009). In addition, we introduce an

original and simple UI that dynamically displays

user input on the sketch, and visually summarizes

the 2D processing results to the user.

By producing simple geometry, our system

strives to minimize the time it may take for modelers

to fix possible artefacts. Compared to similar

systems, our system focuses on the production of

professional-grade geometry and topology details

(Patnode, 2012), which includes the combination of

body parts to produce accurate edge flow around

joints.

Finally, we introduce a novel and simple

approach in 2D outline smoothing, 2D skin-skeleton

mapping, limb axis correction, depth estimation and

mesh smoothing.

2 RELATED WORKS

2.1 Overview

Since the mid-1980s, SBM systems, including

single-view (Alexe, Gaildrat and Barthe, 2004;

Karpenko, Hughes and Raskar, 2002), multiple-view

(Zhou et al., 2016), and data-driven systems,

emerged to help 3D modelers by creating automatic

sketch interpretation systems (Kazmi, You and

Zhang, 2014). Teddy (Igarashi, Matsuoka and

Tanaka, 2007), a single-view system, is considered

the pioneer of interactive sketching UIs that focus on

3D free-form design (Kazmi, You and Zhang, 2014),

although it produced only simple models and

provided limited features. Other SBM systems that

further contributed to simple interactive free-form

modeling techniques include SmoothSketch

(Karpenko and Hughes, 2006) and FiberMesh

(Nealen et al., 2007).

Another important single-view system is

introduced by Gonen & Akleman (2012), which

constructs a “mostly” quad-based model based on

curve partitioning and classification of a sketch. On

the other hand, Rivers et al. (2010) propose a notable

multiple-view system, which focuses on modeling

man-made objects. Viana et al. (2014) propose

another notable approach to create 3D models from

multiple-view concept arts, which focuses on

producing high quality objects at a low cost.

There are many data-driven systems, which use

databases of 3D templates to construct the final 3D

model (Johnston et al., 2015; Mao, Qin and Wright,

2009). The most relevant data-driven system, to our

approach, is the Virtual Human Sketcher (VHS)

(Mao, Qin and Wright, 2006), an established SBM

system that produces simple human-shaped

geometry. However, VHS differs from our approach

because it relies on user interaction in each phase of

the development. In addition, although the geometry

is simple and flexible, the generic template restricts

users to only one character style. VHS, as

demonstrated by the authors, is suitable for low-

quality crowd modeling but not for main characters

in games or movies.

2.2 Single-view, No-database Systems

As our method is a single-view, non-data-driven

system, this section covers the current state-of-the-

art in this specific subcategory of SBM.

A notable system that focuses on constructing

3D characters from single-view images is by

Buchanan et al. (2013). This system reads a concept

image of a character and extracts a 2D skeleton from

the character using an adapted smoothing and

merging algorithm. For the 3D mesh, the system

creates a set of arcs, where each is based on a point

on the outline and a center point on the skeleton.

Finally, a cross-section is extruded along the

skeleton to create the 3D mesh surfaces.

The system introduced in Buchanan et al.’s work

(2013) is a novel approach that produces 3D

characters from single-view concept images.

However, like Johnston et al.’s research (2015), this

system relies on detailed and colored concept images

to estimate the skeleton and character orientation

more accurately. Therefore, we assume that using

simple reference sketches can produce less accurate

mesh results. Moreover, the 3D “arcs” revolving 180

degrees around the skeleton create a different

topology than the standard topology used in game

and movie characters.

Bessmeltsev et al. (2015) created a system that

processes all types of cartoon characters. Their

system integrates a 3D skeleton to help define the

depth of the character and eliminate the ambiguities

of single-view modeling. The system partitions the

character outline, into body parts, based on the

skeleton’s joint-connected bones. The system further

ensures that the symmetry of the character is

maximized. It then creates 3D surfaces by revolving

each surface about its 3D axis. The final 3D model

aims to conform to the character outline, and create

a continuity-persistent model. This system avoids

the disadvantages that Buchanan et al.’s (2013)

system faces, by manually producing the skeleton

(Bessmeltsev et al. 2015). Moreover, it more

accurately estimates the depth and the occluded parts

Creating 3D Human Character Mesh Prototypes from a Single Front-view Sketch

249

of the character based on the 3D skeleton. However,

while the algorithm creates high-quality meshes,

there are topology issues around joined areas, which

the system itself does not resolve, as stated by the

authors.

3 METHODOLOGY

3.1 System Overview

To model a character’s basic geometry, our system

automatically traces the outline of the sketch. It then

displays a simple UI to allow the user to locate

important body joints on the sketch. Then, using the

outline and user input, it automatically derives

additional joints, forming a rough skeleton and

mapping between the outline and skeleton. Finally,

the system estimates the body’s depth, constructs

and attaches body parts and creates a low-polygon

count 3D character mesh. This paper details our

complete process, from the initial input to the final

3D model.

We used Python 2.7, specifically PySide’s QtGui

and QtCore libraries to create a dynamic UI and

process user input. We use the OpenMayaUI module

to open the UI window in Maya, and we use .mel

commands in Python to automatically develop the

3D mesh in Maya’s workspace.

3.2 Sketch & Input Processing

To produce expected results, the character sketch

must meet the following requirements:

1) White or very light-colored background,

and clear outline of the character.

2) A human character in front view, with the

arms extended 45 to 90 degrees away from

the body. Fitted clothing is required.

3) Character’s hair is chin-length or shorter.

No loose garment, gear or other objects.

To extract the character outline, the system

converts the image to binary by finding the

foreground and background pixels. Next, like

Buchanan et al. (2013), we use Potrace (Selinger,

2003) to determine the outline of the character

sketch. Then, we apply a pixel-direction algorithm,

inspired by Lin & Wang’s (2011) feature detection

method, to reduce pixel tracing roughness.

Our UI displays the character image with a

visible traced outline (Fig. 1). Users are required to

input a total of 10 points, locating the neck, biceps,

wrists, hipbone and ankles. The system organizes the

user-specified points in a data structure that

represents a rough skeleton for the character. With

point type recognition, the system only needs to

distinguish between the two points of the same level

(Fig. 2).

Figure 1: A screenshot of the UI.

Figure 2: The outline result (in green) of the character,

with labeled Input Points (User Perspective).

Character Body Sectioning

To add skin-skeleton mappings to the arm (Fig. 3),

the system divides the arm axis into s equally sized

sections by placing s-1 midpoints on the arm axis

from the wrist to the bicep. After that, like the bicep

and wrist points, the system extends two lines

perpendicular to the arm axis from each midpoint,

towards the outline, to find the intersection points,

and then centers each midpoint. The arm axis is no

longer a straight line after the adjustment of

midpoints.

As the hip-leg relation is not equivalent to the

bicep-arm relation, we process the legs differently.

First, the system creates a leg axis from the hip to

the ankle points. Then, it extends horizontal lines

from the ankle point on both sides. Then, the system

tries to locate the top middle point between the legs,

the crotch point. To locate the crotch point, the

system uses the inner leg curve that starts at the right

intersection point of the horizontal line extended

from the ankle point of the left leg and ends at the

left intersection point of the horizontal line extended

from the ankle point of the right leg. Following this,

the system finds the crotch point, the highest pixel

GRAPP 2018 - International Conference on Computer Graphics Theory and Applications

250

on the inner leg curve, or the median of the highest

pixels.

Figure 3: Skin-skeleton mapping of arm points. The

system extends a line (red) from the bicep to the wrist,

forming the arm axis. It then extends lines (blue)

perpendicular to the arm axis, towards the outline, and

looks for the intersection points (purple) with the skin.

Finally, bicep and wrist points are centered between the

two purple intersection points.

Using the crotch point, the system locates the

pelvis corners. Pelvis corners are two points, one on

each side of the crotch point, on the inner leg curve,

horizontally furthest from the crotch point and lower

by 1 pixel. Using those pelvis corners, the system

creates the thigh line on each leg, which starts at the

leg’s pelvis corner and ends at the intersection point

of the outer side of the leg.

When the initial limb axis fails to fit in the

corresponding limb due to the limb’s size or angle,

our algorithm automatically adjusts the axis when it

fails to find an intersection point for every

perpendicular line extended from the axis midpoints.

Fig. 4 demonstrates the automatic axis correction

process.

Figure 4: Limb Axis Correction Process. Top: some

midpoints (blue) have intersection points (purple) from

one side only because of unfitted axes. Middle: using

single-sided intersections, the system extends

perpendicular lines (green) towards the opposite side to

find the second intersection. Bottom: midpoints are

centered.

To section the neck, the system finds

intersections between the two neck points and the

outline. Our algorithm calculates the neck width by

vertically traversing the neck to retrieve the smallest

horizontal diameter in that section, forming a middle

neck line with that width. After that, to determine

the width and height of the head, the system applies

two different approaches (see Fig. 5). The first

approach calculates the head’s height h

from the

sketch and derives the width w

, which is equal to

two-thirds of h

. The second approach calculates the

head’s width w

from the sketch and derives the

height h

as follows:

(1)

The system uses the values of the first approach

by default, unless w

is smaller than w

. As seen in

Fig. 5, h

is the length of l, the vertical line between

the head tip and the neck top point. In addition, w

the smallest horizontal diameter of the middle

section of the head. Having calculated dimensions of

each approach, the system uses values of the second

approach if and only if w

is smaller than w

. In

addition, if the second approach is used, a new head

tip is calculated and a new vertical line l is formed to

match the new height of the head.

Finally, unlike other body parts, vertical head

sectioning, using eight lines, does not depend on

intersection lines but on standard head proportions

that are used to approximate the widths of these

lines.

Figure 5: Head Dimensions. For 1

approach: h1 is the

length of l, the line between the head tip and the neck top

point. For 2

approach: w2 is the smallest diameter found

in the middle section of the head.

Creating 3D Human Character Mesh Prototypes from a Single Front-view Sketch

251

Figure 6: The 1

method of forming the clavicle line. The

system searches for the lowest pixels in the outline portion

(green) between c

& n

, and between c

& n

. The clavicle

points (purple) are the lowest, and closest to m, pixels

forming the clavicle line.

For the torso, the system first locates the

underarm curve using an automatic body feature

extraction algorithm (Lin and Yang, 2011). Then, as

seen in Fig. 6, guided by the neck section and

underarm curve, the system estimates the character’s

clavicle line. In the cases where this approach

doesn’t work, the system forms a clavicle line that

has the same width as the neck bottom line. Next, to

prepare for sufficient 3D geometry in the shoulder

area, our algorithm adds a shoulder line, between the

clavicle and chest lines, with a width equal to the

average width of the two lines.

Figure 7: A complete 2D blueprint of the sketched

character.

For the rest of the torso, the system inserts five

equally spaced horizontal lines between the chest

and thigh lines and forms the area of each shoulder

with three lines placed between the upper torso and

the arm. Fig. 7 presents the final 2D blueprint of all

the section lines that the system automatically

derives using the 10 user-input points.

3.3 3D Construction

Our 3D construction process is divided into depth

estimation, mesh construction and mesh smoothing.

We implemented some depth constraints that are

suitable for human characters, and maintained the

flexibility to create nonrealistic body proportions.

Similar to its width and height, we derive the

head’s depth using relatively standard proportions,

which we adjusted based on empirical tests with a

variety of character sketches, and it is equal to

seven-eighth the height of the head. For the limbs,

our algorithm simply makes the depth of each limb

section line equal to its width, creating circular arms

and legs. This decision is influenced by the box

modeling process, where limbs are blocked out

using cylinders (Villar, 2014). In addition, for

simplicity, the algorithm keeps the neck middle and

bottom lines circular, but locally manipulates them

during the smoothing process.

To model the torso, the chest depth d

is first

calculated based on empirical tests to estimate a

suitable torso depth, and is influenced by NASA’s

anthropometric data (Churchill et al., 1978). In

addition, d

is constrained by width w

of the chest

line. d

is first assigned the greater value among the

two following expressions:

* 1.05 (2)

+ d

(3)

where d

, in (2), is the depth of the neck bottom. In

(3), d

and d

are the depths of the bicep lines of the

left and right arms, respectively. If d

is greater than

, then it is adjusted to be:

(4)

Next, the system calculates the depth of the clavicle

and shoulder lines based on the quadratic Bezier

function as follows:

= (1 - t)

+ 2(1 - t)tp

+ t

(5)

such that

t = (m

- m

)/(m

- m

) (6)

In equation 5, p

is the bottom neck depth, p

the chest depth, and p

-1. In equation 6, m

is the

midpoint of the upper torso line in process, m

is the

midpoint of the neck bottom line, and m

is the

midpoint of the chest line. To estimate the lower

torso depth, the system first calculates the hip depth

, which is the average depth of the thigh lines. In

addition, it retrieves m

, the midpoint of the thigh

lines. After that, for each lower torso line, the

system calculates the depth d

by linearly

interpolating between d

and d

as follows:

= (r * (d

- d

)) + d

(7)

such that

r = (m

- m

)/(m

- m

) (8)

GRAPP 2018 - International Conference on Computer Graphics Theory and Applications

252

In equation 8, m

is the midpoint of the torso line in

process, and m

is the midpoint of the chest line.

3D Mesh Construction & Smoothing

To create the character’s mesh, the system converts

2D and depth data into a 3D structure that defines

the shape of the character. To do that, the system

uses the relative position and orientation of each

body section line and converts it into a 3D profile

curve, reflecting its calculated depth. Using profile

curves, the system constructs polygon surfaces on

which edge loops are formed to reflect the position

and orientation of those curves. Fig. 8 shows front

and side views of the polygon surfaces.

To complete the head mesh, the system attaches

a new surface, subdivided for suitable edge flow, on

each end of the head. The system then attaches the

head bottom to the neck by bridging between the

neck’s top edge loop and the head bottom’s back

edges, deleting unnecessary surface faces. To

smooth the head top and joined head-neck area, the

system implements a series of Bezier curves.

Figure 8: Polygon surfaces constructed using 2D

coordinates and orientation, as well as depth data.

To connect the torso to the legs, the system

partially bridges between the thigh lines and the

bottom edge of the torso, leaving the middle section

unconnected. When constructing the middle surface,

the system enforces a minimal distance of 0.1

centimeters between the inner leg vertices to prevent

overlapping geometry. After that, the inner legs and

torso are joined by a single surface, subdivided for

suitable edge flow. Finally, the system implements a

series of quadratic Bezier curves (equation 5) to

smooth the geometry of the hips. In addition, the

system curves two edges, using their vertices,

around the hips to follow the joint and muscle

structure.

Connecting shoulders to the torso, the last

process the system automatically applies, is different

as the top shoulder line is already aligned with the

upper torso lines from both sides (Fig. 8). Thus, the

system averagely merges aligned shoulder and upper

torso vertices, eliminating torso side faces bounded

by those vertices.

The result is a connected mesh that is smoothed

for better appearance and enhanced geometry. Our

algorithm strives to follow the standard and

recommended guidelines in modeling human

characters, producing suitable geometry and

topology. Fig. 9 shows a complete 3D mesh of the

character. In addition, Fig. 10 shows a sample of two

sketches along with the system-generated mesh.

Note the cartoon-like proportions (matching the

sketches) as well as the proper edge flow around the

shoulders, hips and neck.

Figure 9: 3D character mesh. From left to right, the first

two are in front view. The third and fourth are in side and

perspective views, respectively.

Figure 10: Sample of a system-generated mesh.

4 EXPERIMENTS AND RESULTS

4.1 User Study

We conducted a user study to test our system with a

variety of sketched characters and collect feedback

from 3D artists. We asked participants to bring

sketches that met the requirements listed earlier in

this paper. After a short demonstration, we

proceeded with the study as follows:

Using their character sketches, participants used

the UI to input the 10 required points on the

character, and had the system automatically produce

the 3D character mesh. The system logged the time

durations of user interaction, 2D sketch processing

and 3D mesh construction, and the users completed

a post-study questionnaire.

We had a total of 15 participants, all below the

age of 40. The system succeeded in producing a

mesh with 14 out of the 15 participant sketches.

Users rated their level of experience in 3D modeling

generally, and in character modeling particularly. In

3D modeling, 7% said they never modeled before,

Creating 3D Human Character Mesh Prototypes from a Single Front-view Sketch

253

29% rated themselves as beginners, 21% as

intermediate, 36% as advanced, and 7% as expert.

On the other hand, in character modeling, 14% said

they never modeled characters before, 43% were

beginners, 29% intermediate, and 14% advanced.

Advanced artists estimated that a similar mesh

would take them between 30 minutes and an hour to

create.

User Evaluation

Regarding the sketch specifications required by the

system, 29% of the users said the specifications were

not enough, suggesting that an added side-view

sketch would create better depth results. 57% said

the amount was enough, expressing that the

requirements were “reasonable” and

“straightforward”. 14% said “a lot” because they

limit the flexibility to accept additional details like

“weapons or other pieces of armor”. On the other

hand, 7% of users said that the 10-point input was

too little, 86% said it was enough, and 7% said it

was too much, as input points could be mirrored

when using symmetric characters. When asked to

rate the learning curve and ease of interaction with

our UI, 100% of users agreed or strongly agreed that

the system was easy to learn and to use. All users

also agreed or strongly agreed that the system

generated the mesh quickly.

86% of the users said that they would use the

generated mesh to build their 3D character. Overall,

comments about the system results were positive.

Users’ description of the system included “very

useful”, “quick”, “efficient”, and “accurate size

estimation”. Their suggestions included “more

flexibility” and “more depth control”.

4.2 Discussion

The results of the user study show promising

feedback regarding our system. All users agreed that

the UI is easy to understand and interact with. Users

stated that it was “straightforward” and “very easy to

manipulate”, although one user suggested changing

the “wizard style interface”, and another suggested

that it would be easier to “pre-generate” adjustable

points on the sketch. Users also agreed that the

system generated the mesh quickly. Most users also

highly rated the geometry and topology of the mesh,

and expressed their willingness to use the mesh to

construct the character model. On the other hand, a

few users criticized the limitations of a system that

does not recognize additional objects like hats or

bags, and is less accurate with non-fitted garment.

Our system generated the mesh in an average of

11.5 seconds, including 2D and 3D processing. The

approximate average time of user interaction was 2

minutes and 13 seconds. We did not find a

significant correlation between manual construction

time and user experience, and it was interesting to

see that the approximated manual time to block out

the basic geometry of the character does not

decrease as modelers become more experienced.

Finally, part of our research questions was to

determine if users value the speed of an automatic

SBM system. Users’ preference in using an

automatic system versus manual modeling did not

greatly change when the system hypothetically lost

its speed advantage. Novice modelers (beginners)

leaned towards automatic modeling regardless of the

time it takes, which supports Alhashim’s (2015)

statement about how novice modelers find it less

tedious to work with existing meshes rather than

starting from scratch. On the other hand,

intermediate and advanced modelers were more

likely to base their decision to use the automatic

system on its capabilities and the specific details it

produces in a mesh.

5 CONCLUSIONS

In this paper, we proposed a system that

automatically creates the basic geometry of a 3D

human character using a single-view sketch and

minimal user input. Our goal was to create an

algorithm that conforms to the current modeling

techniques as well as the geometry and topology that

professional modelers recommend. User evaluation

of the mesh indicates that our algorithm produces

good and usable meshes. The user study conducted

to evaluate this system resulted in positive feedback

from both beginner and intermediate users who

found it helpful and quick to start up the process.

5.1 Limitations & Future Work

Future work could improve the method that locates

the clavicle line, as we encountered issues when the

character’s arms were not fully extended in a T-

pose. In addition, to reduce the complexity of

adjusting head measurements, our system produces

simple geometry that is easy to manually fix.

However, for more accurate results, the system

could use internal strokes of the sketched character

to trace the hairline and estimate the hair volume.

In conclusion, despite the limitations of our

system, results and feedback indicate that our system

produces suitable results for non-advanced modelers

to speed up the modeling process, using an intuitive

GRAPP 2018 - International Conference on Computer Graphics Theory and Applications

254

and easy to use UI. 3D modeling students showed

interest in using the system, and modeling experts

expressed the usefulness of the system to beginners.

This proves that our system contributes is a step

towards building more familiar and supported SBM

system.

ACKNOWLEDGEMENTS

This project was funded by The Interactive and

Multi-Modal Experience Research Syndicate

(IMMERSe).

REFERENCES

Alexe, A., Gaildrat, V. and Barthe, L., 2004, November.

Interactive modelling from sketches using spherical

implicit functions. In Proceedings of the 3rd

international conference on Computer graphics,

virtual reality, visualisation and interaction in Africa

(pp. 25-34). ACM.

Alhashim, I., 2015. Modeling and Correspondence of

Topologically Complex 3D Shapes.

Andre, A. and Saito, S., 2011, August. Single-view sketch

based modeling. In Proceedings of the Eighth

Eurographics Symposium on Sketch-Based Interfaces

and Modeling (pp. 133-140). ACM.

Bessmeltsev, M., Chang, W., Vining, N., Sheffer, A. and

Singh, K., 2015. Modeling character canvases from

cartoon drawings. ACM Transactions on Graphics

(TOG), 34(5), p.162.

Churchill, E., McConville, J.T., Laubach, L.L., Erskine,

P., Downing, K. and Churchill, T., 1978.

Anthropometric source book. Volume II: handbook of

anthropometric data.

Buchanan, P., Mukundan, R. and Doggett, M., 2013, July.

Automatic single-view character model reconstruction.

In Proceedings of the International Symposium on

Sketch-Based Interfaces and Modeling (pp. 5-14).

ACM.

Celebi, M.E., 2011. Improving the performance of k-

means for color quantization. Image and Vision

Computing, 29(4), pp.260-271.

Cook, M.T. and Agah, A., 2009. A survey of sketch-based

3-D modeling techniques. Interacting with computers,

21(3), pp.201-211.

Dargie, J., 2007. Modeling techniques: movies vs. games.

ACM SIGGRAPH Computer Graphics, 41(2), p.2.

Entem, E., Barthe, L., Cani, M.P., Cordier, F. and Van de

Panne, M., 2015. Modeling 3D animals from a side-

view sketch. Computers & Graphics, 46, pp.221-230.

Gonen, O. and Akleman, E., 2012. Sketch based 3D

modeling with curvature classification. Computers &

Graphics, 36(5), pp.521-525.

Igarashi, T., Matsuoka, S. and Tanaka, H., 2007, August.

Teddy: a sketching interface for 3D freeform design.

In Acm siggraph 2007 courses (p. 21). ACM.

Johnston, A., Carneiro, G., Ding, R. and Velho, L., 2015,

November. 3-D Modeling from Concept Sketches of

Human Characters with Minimal User Interaction. In

Digital Image Computing: Techniques and

Applications (DICTA), (pp. 1-8). IEEE.

Karpenko, O.A. and Hughes, J.F., 2006, July.

SmoothSketch: 3D free-form shapes from complex

sketches. In ACM Transactions on Graphics (TOG)

(Vol. 25, No. 3, pp. 589-598). ACM.

Kazmi, I.K., You, L. and Zhang, J.J., 2014, August. A

survey of sketch based modeling systems. In

Computer Graphics, Imaging and Visualization

(CGIV), 2014 11th International Conference on (pp.

27-36). IEEE.

Lin, Y.L. and Wang, M.J.J., 2011. Automated body

feature extraction from 2D images. Expert Systems

with Applications, 38(3), pp.2585-2591.

Mao, C., Qin, S.F. and Wright, D., 2009. A sketch-based

approach to human body modelling. Computers &

Graphics, 33(4), pp.521-541.

Mao, C., Qin, S.F. and Wright, D.K., 2006. Applying

scenarios in user-centred design to develop a

sketching interface for human modelling and

animation. Eurographics.

Nealen, A., Igarashi, T., Sorkine, O. and Alexa, M., 2007.

FiberMesh: designing freeform surfaces with 3D

curves. ACM transactions on graphics (TOG), 26(3),

p.41.

Olsen, L., Samavati, F.F., Sousa, M.C. and Jorge, J.A.,

2009. Sketch-based modeling: A survey. Computers &

Graphics, 33(1), pp.85-103.

Patnode, J., 2012. Character modeling with Maya and

ZBrush: professional polygonal modeling techniques.

CRC Press.

Rivers, A., Durand, F. and Igarashi, T., 2010. 3D modeling

with silhouettes (Vol. 29, No. 4, p. 109). ACM.

Selinger, P., 2003. Potrace: a polygon-based tracing

algorithm. Potrace (online), http://potrace.

sourceforge. net/potrace. pdf (2009-07-01).

Viana, R.D., Nascimento, E.R. and Ferreira, R.A., 2014.

On the Development of a Fully Automatic

Methodology to Create Smooth Mesh for 3D Models

from Concept Arts.

Villar, O., 2014. Learning Blender: a hands-on guide to

creating 3D animated characters. Addison-Wesley

Professional.

Zhou, X., Chen, J., Chen, G., Zhao, Z. and Zhao, Y., 2016.

Anthropometric body modeling based on orthogonal-

view images. International Journal of Industrial

Ergonomics, 53, pp.27-36.

Creating 3D Human Character Mesh Prototypes from a Single Front-view Sketch

255