12 July/August 2011 Published by the IEEE Computer Society 0272-1716/11/$26.00 © 2011 IEEE

Applications Editor: Mike Potel

Digitally Interpreting Traditional Folk CraftsTurlif VilbrandtDigital Materialization Lab, Japan

Carl VilbrandtDigital Materialization Lab, Chile

Galina Ivanovna PaskoBritish Institute of Technology and E-commerce

Cherie StammUformia

Alexander PaskoBournemouth University

Digitally preserving and interpreting cul-tural heritage has attracted considerable attention in the computer graphics, geo-

metric-modeling, VR, and general computer sci-ence communities. When digitally preserving an object, it’s important to consider its geometric shape, especially for 3D physical artifacts in vari-ous crafts, sculpture, and architecture.

Any traditional craft is a living tradition. Part of what keeps the tradition alive is the masters who are familiar with the craft’s essential technology, which is rarely presented in written form. The tra-dition of passing craft-making processes through successive generations of masters and the continu-ous reproduction of the items helps crafts preserve their shapes. As the number of masters decreases, technologies fade, and crafts lose economic ground.

Digital technologies for preserving heritage of-fer the chance to capture and preserve traditional crafts. One such technology, digital fabrication, continues to gain attention by moving beyond its traditional applications. Where cultural heritage and digital fabrication meet, it’s possible to pre-serve the living traditions of crafts, while offering new approaches to their design and production. Here, we present our long-term applied-research project to develop a mathematical basis for digital preservation, software tools that combine interac-tive and automatic design, and technology for ap-plying digital fabrication to traditional crafts.

Shape RepresentationOur project encompasses these R&D directions:

■ modeling shapes and making parametric fami-lies of models of representative craft items (this lets us generate samples of a model in different sizes, widths, height ratios, and so on);

■ producing 3D virtual craft objects and present-ing them on the Web;

■ documenting traditional materials and technology;■ developing interactive design tools for modeling

new items (this is a radical step in developing special CAD tools for modeling a craft’s shapes and material properties);

■ applying existing digital-fabrication machines to produce 3D physical objects from computer models;

■ adapting or designing digital-fabrication tools to manufacture craft items on a desktop; and

■ creating an Internet-based community and e- commerce activity using interactive CAD, virtual-object presentation, and on-demand fab-rication of items.

How we implement the directions depends greatly on the mathematical shape representation. The dominant approach for representation of cul-turally valuable objects is boundary representation. However, we found it inadequate for our purposes (for more information, see the sidebar).

IEEE Computer Graphics and Applications 13

Instead, in our approach we use function repre-sentation (FRep)1 and employ real measurements, constructive modeling, and digital fabrication. This lets designers interactively create constructive 3D object models with some degree of automation rather than using completely automatic mesh re-construction from input data.

Such a modeling technique provides a better un-derstanding of the physical relationships of the ac-tual structure’s components and of the processes used to create the original object. The 3D model can replicate the actual construction of the origi-nal object itself, including normally imperceptible features (such as interior bracketing). The model can then be deconstructed to reveal such hidden features. This approach is especially valuable if the real object has been lost, destroyed, or damaged. The goal is to create as complete a 3D model of the object as possible—that is, a model represent-ing the artifact’s internal structure, design logic (showing how components are interconnected or layered), and shape-construction history, as well as time-dependent aspects and other parametric dependencies.

FRep represents a 3D object as a continuous func-tion of point coordinates, F(x, y, z) ≥ 0. A point be-longs to the object if the function is nonnegative at the point. The function is zero on the object’s entire surface (normally called an implicit surface) and

is negative at any point outside the object. Design-ers can easily parameterize the function to support modeling of a parametric family of objects.

Valery Adzhiev and his colleagues introduced the HyperFun language to teach FRep modeling and to facilitate its practical use.2,3 HyperFun is a good tool for preserving and digitally fabricating cultural-heritage objects for many reasons, includ-ing these:

■ It has an open and simple textual format. ■ It has a well-defined mathematical basis. ■ It supports constructive, parameterized, and multidimensional models.

■ It’s supported by free and open source modeling and visualization software.

We reported our project’s earlier stages—modeling lacquerware artifacts and Web presentation of the models—elsewhere.4 Here, we concentrate on au-tomating fabrication of traditional crafts and on modeling and manufacturing new designs and forms. Our test applications involved crafts from two cultural backgrounds: Japanese lacquerware and Norwegian wood carvings.

Japanese LacquerwareShikki (traditional Japanese lacquerware) artisans take thin pieces of wood, paint them different

B oundary representation (BRep), which employs polygo-nal meshes and parametric surfaces, has worked well

for interactive visualization. However, BRep poses quantita-tive and qualitative problems for applications involving the physical structure and techniques related to an artifact, rep-resentation of parametric families of artifacts, Web-based modeling, long-term archiving (for hundreds of years), or physical reproduction through digital fabrication.

Size and Processing TimeModerate-size surface-based models containing high-quality details can include so many polygons that modern graphics hardware either has difficulty rendering them or can’t render them. File size can become a serious problem for Web-based model presentation and manipulation.

Validity and PrecisionTraditional BRep-based CAD models can present problems, such as surface cracks, self-intersections of polygons, addi-tional false polygons left over from modeling, and inverted normal orientation appearing in meshes reconstructed from scan data. Generally, BRep models aren’t exact; they only approximate the modeled geometry, with limited precision.

Parameterization and OperabilitySupport for model generation with variable parameters is important for classifying artifacts and determining author-ship. BRep has limited or no support for parameterization. When users change a model’s parameters, they must regen-erate BRep models using a separate, high-level procedure. They might need to apply further operations such as off-sets, blends, and shape deformations and metamorphoses, for which BRep modeling systems provide limited or no support.

ManufacturabilityBRep models grow dramatically and become difficult or impossible for current hardware systems to visualize or cross-section, as is required by many digital-fabrication systems and processes. Although current digital-fabrication systems have limited resolution, they’ve achieved ever-greater accuracy; this trend should continue. Even so, objects often have defects or missing sections. This is due largely to the complexity of creating proper cross sections from the standard stereolithography format (known as STL), which represents the object’s surface as a set of disconnected triangles.

Boundary Representation and Its Limitations

14 July/August 2011

Applications

colors, assemble them, and cover them with urushi, a natural lacquer. As with other traditional crafts, shikki is becoming a fading industry owing to cheap plastic production. So, preserving it is critical.

We first created 3D computer models of shikki using HyperFun (see Figure 1). We then made po-lygonal approximations of object surfaces using the HyperFun Polygonizer software and exported the generated mesh to VRML (Virtual Reality Modeling Language) format.

We created the Virtual Shikki website (http://hyperfun.org/wiki/doku.php?id=apps:shikki) to showcase our models of a lacquerware box, tray, cup, stand, sake pot, and a full sake set. A Hy-perFun model is available for each object on the website. Each image is linked to the corresponding VRML model, which you can download and visual-ize using any VRML viewer, such as CosmoPlayer.

Because tooling is an important part of fab-rication, we experimented with different digital- fabrication equipment to produce the items we de-signed. Many rapid-prototyping machines fabricate objects that are best suited for viewing, not han-dling, because the materials are too fragile. Also, depending on the object, fine detail and finishing might not be possible. For example, Figure 2 shows spoons we fabricated using a paper-laminating Kira solid-center rapid-prototyping machine.

We’ve experimented primarily with the Roland Modela MDX-20, a low-cost, easy-to-use desktop scanner and milling machine. Unlike 3D printers, the MDX-20 supports hardwoods and can produce fine detail and design. Figure 3 shows a wooden spoon we fabricated with the MDX-20, using the model in Figure 1b.

We’re expanding from traditional lacquerware objects to other types of objects. For example, we created original spoon and bonbonniere (candy dish) designs and tested an experimental fabri-cation technique (see Figures 4 and 5). As Figure 5b shows, when fabricating the bonbonniere lid’s top and bottom, we used wax to hold the pieces in place.

Norwegian Wood CarvingsWe used a traditional wood carving from Norway’s Lyngen region (see Figure 6). This carving is con-sidered a symbol of the local community; we de-cided to create a 3D-relief version of it to use to produce jewelry and other things.

We measured the original carving and modeled the object using HyperFun to produce a casting mold. The 3D model is a composite of relief mod-els of several parts of the carving such that it’s suitable for casting (see Figure 7). We generated the relief models using polygon-to-function con-version5 and applied the models to the outlines of the original carving’s elements. We fabricated the cast’s form out of modeling wax on the MDX-20

(a)

(b)

Figure 1. Modeling Japanese lacquerware. (a) Example lacquerware spoons. (b) A spoon modeled in HyperFun. Our goals were to post realistic models on our Virtual Shikki website (http://hyperfun.org/wiki/doku.php?id=apps:shikki) and fabricate objects based on the models.

Figure 2. Spoon models fabricated on a rapid-prototyping Kira solid-center machine using laminated paper. This method couldn’t produce finely detailed and finished objects.

Figure 3. A spoon fabricated with a Roland Modela MDX-20 milling machine, using the model in Figure 1b. The MDX-20 is inexpensive and easy to use on hardwood, wax, and other materials.

IEEE Computer Graphics and Applications 15

(see Figure 8a). (The total milling time to finish the object’s surface was 18 hours.) We used the wax model to make a mold—a negative cast—from gypsum (see Figure 8b).

Creating a physical representation of a Hyper-Fun model using the MDX-20 was a delicate, time-consuming process. We started by creating a stereolithography (STL) file from the HyperFun model. We imported this file into the Modela Player milling software. Then, we scaled the STL model to fit the limitations of the MDX-20’s bed and the target object’s size.

The MDX-20 milling process generally takes three steps for each side that’s milled:

1. Plane the material into a flat surface, parallel to the machine’s bed.

2. Make a rough milling, removing most of the

material and leaving a rough layer on top that approximates the object’s shape.

3. Make a finishing pass, removing the excess material and providing a finished surface.

Although we could have taken the resulting molds to a professional jeweler, we decided to produce the jewelry ourselves, using special clay developed by Art Clay. The clay consists of 1- to 20-micron silver particles, organic binders, and water. During the firing, the organic binders burn away and the silver particles become denser and stronger, resulting in an object that’s 99.9 per-cent silver. After experimenting with the different forms of this material, we found that using the syringe-type clay produced the best results for fill-ing the molds. Following Art Clay’s instructions, we heated the filled molds in an oven for a short

(a) (b)

Figure 4. Designing and fabricating spoons. (a) An untraditional “organic” design, giving the appearance of walnut. (b) New designs we fabricated in wood. These are examples of using traditional designs to create new modern designs.

(a) (b)

Figure 5. Designing and fabricating a bonbonniere (candy dish). (a) The bonbonniere’s design. (b) The bonbonniere lid, fabricated in wood. When fabricating the lid’s top and bottom, we used wax to hold the pieces in place.

16 July/August 2011

Applications

period to dry the clay. After removing the silver objects from the molds, we fired some in a profes-sional kiln and the rest in an inexpensive portable kiln, each with similarly successful results. Polish-ing the objects gave them a professional-quality look (see Figure 9).

We used this process to develop a curriculum to teach high-school design students how to design and fabricate their own creations. The students

learned the entire process, starting with design-ing their jewelry on paper. After a few sessions in 3D software (for example, HyperFun), they transferred their designs from paper to digital 3D models. Following our process, they made molds of their designs using the MDX-20. Next, they used the Art Clay material to fill their molds (see Fig-ure 10a) and dried, fired, and polished the cast-ings. The students then put the finishing touches on their designs using traditional jeweler’s tools. Each student left the class with a digital model and a piece of silver jewelry he or she created from scratch (see Figure 10b).

System Evaluation and DiscussionUntil recently, we had several problems using FRep for cultural-heritage applications. We’ve dealt with these problems to develop a usable modeling, rendering, and fabrication system.

Real-Time Direct RenderingThe main disadvantage with FRep modeling was its time-consuming function evaluation and slow object rendering, which hindered interactive ap-plications. Typically, an FRep object must be ap-proximated by a polygonal mesh, which we could then render using graphics hardware. Recently, Oleg Fryazinov and his colleagues introduced a method for direct ray casting and ray tracing of FRep objects using revised affine arithmetic and its extensions.6 Implementing this method on a GPU provided real-time direct rendering (four frames per second) of the entire sake set from the Virtual Shikki project.

Interactive ModelingFRep’s fast direct rendering provided a basis for implementing an interactive modeling system, which avoided tedious coding in HyperFun. Our team have implemented prototype interactive modelers as Autodesk 3ds Max and Maya plug-ins. We provided different operation modes for users, such as direct object manipulation, a command-line interface, and a tree interface (see Figure 11).

Constructive modeling usually isn’t automated; it takes time and requires some 3D-modeling skill. So, we examined several automated-modeling ap-proaches, such as fitting a parameterized FRep template model to a cloud of data points and ex-tracting the model’s logical structure in the form of a construction tree using genetic algorithms.7

Web PresentationAt first, VRML seemed a natural choice for pre-senting our 3D virtual objects on the Web.4 How-

Figure 6. A line drawing of the original wood carving from Norway’s Lyngen region. We created a 3D-relief version of it to use to produce jewelry and other things.

Figure 7. A 3D HyperFun model of the original carving. The model is a composite of relief models of several parts of the carving such that it’s suitable for casting.

IEEE Computer Graphics and Applications 17

ever, it has well-known drawbacks, such as huge data files and therefore long downloading times. For example, an average-sized VRML file is 100 to 500 Kbytes. So, we should consider more concise Web3D formats such as HyperFun in the future. In VRML, the full sake set is 4.5 Mbytes; the Hyper Fun models for all the lacquerware items are less than 5 Kbytes. So, we conclude that, because HyperFun reduces everything to a mathematical equation or procedure with a high compression level, users should consider it for a lightweight network pro-tocol. A Web browser plug-in let us use HyperFun for our Web presentations.3 With the plug-in, we can transfer the small HyperFun models and un-fold a polygonal mesh or support direct rendering suitable for interactive visualization.

Direct FabricationWe manufactured several of our FRep models us-ing the MDX-20 with standard input in the form of STL files. We imported the STL files into the machine’s standard software package, checked the imported models for surface defects, and gener-ated machine paths on the basis of layered slices. During manufacturing, the STL format created many issues for slicing, path planning, and fabri-cating some of the models’ fine features.

A much better approach is to directly fabricate the FRep model without using poor intermediate formats, as Turlif Vilbrandt and his colleagues have proposed.8 One possibility is to produce a raster im-age for each manufacturing layer at the machine resolution, which is acceptable input for some ma-chines. Another is to directly control digital fabri-cation, including the tool’s motions and material deposition. We plan to develop and experiment with multimaterial 3D printers to reproduce sur-face painting and other craft features.

(a) (b)

Figure 8. Changing the 3D model into a negative cast (a mold). (a) A wax model fabricated with the MDX-20. (b) The negative cast made from gypsum, using the wax model. The total milling time to finish the wax model’s surface was 18 hours.

Figure 9. A finished piece of silver jewelry depicting the Norwegian rider carving. We produced this object ourselves, using a special clay from Art Clay consisting of silver particles, organic binders, and water.

(a) (b)

Figure 10. Students from the Breivang School of Design in Tromsø, Norway, used our approach to design and fabricate silver jewelry. (a) Students filling their molds with silver clay. (b) Some finished jewelry.

18 July/August 2011

Applications

As our results show, inexpensive desktop fabri-cation equipment with the appropriate soft-

ware can help people preserve and support existing craft techniques and create new design and fab-rication approaches. We’ve found digital technol-ogy useful in extending and enhancing traditional processes, without disrupting or replacing them. This means that traditional shapes and forms can be preserved and can continue to develop.

References 1. A. Pasko et al., “Function Representation in

Geometric Modeling: Concepts, Implementation, and Applications,” The Visual Computer, vol. 11, no. 8, 1995, pp. 429–446.

2. V. Adzhiev et al., “HyperFun Project: A Framework for Collaborative Multidimensional FRep Modeling,” Proc. 1999 Eurographics/ACM Siggraph Workshop Implicit Surfaces, Eurographics Assoc., 1999, pp. 59–69.

3. R. Cartwright et al., “Web-Based Shape Modeling with HyperFun,” IEEE Computer Graphics and Applications, vol. 25, no. 2, 2005, pp. 60–69.

4. G. Pasko et al., “Virtual Shikki and Sazaedo: Shape Modeling in Digital Preservation of Japanese Lacquer-ware and Temples,” Proc. Spring Conf. Computer

Graphics, IEEE CS Press, 2001, pp. 147–154. 5. A. Pasko, V. Savchenko, and A. Sourin, “Synthetic

Carving Using Implicit Surface Primitives,” Computer-Aided Design, vol. 33, no. 5, 2001, pp. 379–388.

6. O. Fryazinov, A. Pasko, and P. Comninos, “Fast Reliable Interrogation of Procedurally Defined Implicit Surfaces Using Extended Revised Affine Arithmetic,” Computers & Graphics, vol. 34, no. 6, 2010, pp. 708–718.

7. C.W. Vilbrandt et al., “Cultural Heritage Preservation Using Constructive Shape Modeling,” Computer Graphics Forum, vol. 23, no. 1, 2004, pp. 25–41.

8. T. Vilbrandt et al., “Universal Desktop Fabrication,” Heterogeneous Objects Modelling and Applications, LNCS 4889, Springer, 2008, pp. 259–284.

Turlif Vilbrandt is a senior researcher at the Digital Ma-terialization Lab in Japan. Contact him at turlif@turlif.org.

Carl Vilbrandt is a senior researcher at the Digital Materi-alization Lab in Chile. Contact him at carl@cgpl.org.

Galina Ivanovna Pasko is an associate professor in com-puter animation at the British Institute of Technology and E-commerce. Contact her at gip@pasko.org.

Cherie Stamm is the director of applied research at Ufor-mia. Contact her at cherie@uformia.no.

Alexander Pasko is a professor in computer animation at Bournemouth University. Contact him at apasko@bournemouth.ac.uk.

Contact department editor Mike Potel at potel@wildcrest.com.

Figure 11. An interactive FRep modeler as an Autodesk Maya plug-in (implemented by Denis Kravtsov). Interactive modeling enables exporting HyperFun models, thus avoiding tedious coding.

Visit CG&A on the Web at

www.computer.org/cga