methodologies for the visualization ... - palaeo-electronica · ment, the most popular of which...

Palaeontologia Electronica http://palaeo-electronica.org

METHODOLOGIES FOR THE VISUALIZATION AND RECONSTRUCTION OF THREE-DIMENSIONAL FOSSILS FROM THE SILURIAN HEREFORDSHIRE LAGERSTÄTTE

Mark D. Sutton, Derek E.G. Briggs, David J. Siveter, and Derek J. Siveter

Mark D. Sutton. Department of Earth Sciences, University of Oxford, Parks Road, Oxford, OX1 3PR, U.K. Derek E.G. Briggs. Department of Earth Sciences, Wills Memorial Building, Queen's Road, Bristol, BS8 1RJ, U.K.David J. Siveter. Department of Geology, Bennett Building, University of Leicester, University Road, Leicester, LE1 7RH, U.K.Derek J. Siveter. University Museum of Natural History & Department of Earth Sciences, University of Oxford, Parks Road, Oxford, OX1 3PR, U.K.

ABSTRACT

The dominantly soft-bodied fauna of the Silurian (Wenlock) Herefordshire Lager-stätte of England is preserved as three-dimensional calcitic fossils within calcareousnodules. These fossils are difficult to interpret from study of sections produced by split-ting, and are difficult also to isolate from their matrix. CT scanning of nodules is feasi-ble, but is expensive and unlikely to be able to image the fossils at a sufficiently highresolution. We describe a technique based on serial grinding and digital photographywhich enables the capture of three-dimensional morphological information from thefossils. Computer video files are used to package the resulting datasets for inspection .Three-dimensional reconstructions can be created by a method involving the manualtracing of outlines on each image, but we consider this approach inappropriate in thisinstance. Datasets from grinding are treated instead as volumes, which are renderedeither directly to produce virtual X-ray images, or by the computation of isosurfaceswhich are then visualized by ray-tracing. Rendering is performed from sequentialangles to produce video files of rotating three-dimensional models. The method is illus-trated with reconstructions of a trilobite and graptolite associated with the Hereford-shire soft-bodied fauna, and of the worm-like mollusc Acaenoplax hayae.Reconstructions produced in this way can be retouched easily and manipulated toremove noise and preservational artifacts, to dissect out structures for detailed study,or to combine part and counterpart into a single reconstruction. The approach will pro-vide the primary means of study of the Herefordshire fauna, and has the potential toprovide a method of obtaining and/or communicating morphological information in avariety of palaeontological applications.

KEY WORDS: isosurface, reconstruction, Silurian, soft-bodied, three-dimensional.

Copyright: Palaeontological Association, 22 June 2001Submission: 23 March 2000 Acceptance: 12 March 2001

Sutton, Mark D., Briggs, Derek E. G., Siveter, David J., and Derek J. Siveter, 2001. Methodologies for the Visualization and Reconstruction of Three-dimensional Fossils from the Silurian Herefordshire Lagerstätte. Palaeontologia Electronica, vol. 4, issue 1, art. 1: 17pp., 1MB. http://palaeo-electronica.org/2001_1/s2/issue1_01.htm

MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS

INTRODUCTION

The non-biomineralized fauna of theHerefordshire Lagerstätte (Briggs et al .1996) represents a rare opportunity tosample the true taxonomic compositionand palaeoecology of a Silurian benthiccommunity. Fossils occur in spherical tosubelliptical calcareous nodules found in alocally thick bentonite deposit in the Coal-brookdale Formation, Wenlock Series .They vary in size from under a millimetreto several centimetres, with worms,sponges, and non-biomineralized arthro-pods dominating the assemblage. All taxaare preserved in three dimensions, with noindication of post-mortem disturbance orcompression. While some fossils preserveorganic material (e.g., the graptolites) andothers siliceous skeletons (e.g., the radi-olarians), for the most part specimensconsist of dark-coloured sparry calcite

shapes within a lighter-coloured matrix(Briggs et al. 1996; see also Orr et al.2000). Few unambiguous internal featuresof the organisms are known, but externaldetails are commonly reproduced with adegree of fidelity sufficient to preserve finespines 30 µm or less in diameter (Figure1).

All taxonomic and palaeoecologicalstudy of the Herefordshire biota must beunderpinned by careful resolution of themorphologies of the fossils preserved.However, while three-dimensional preser-vation is clearly richer in morphologicalinformation than the two-dimensional ornear two-dimensional modes of most otherKonservat-Lagerstätten (e.g., the BurgessShale, see Allison and Briggs 1991), itintroduces special problems. Surfacesexposed by splitting nodules are essen-tially randomly oriented with respect to theorganism and show only those structures

Figure 1. Detail of fine spines preserved on margin of Acaenoplax hayae Sutton et al. (2001), from the HerefordshireLagerstätte; OUM C.29503b.

2


that happen to be intersected by the split.Structures of fossils are thus easily over-looked or misinterpreted. Furthermore, it isfrequently difficult to distinguish the genu-ine termination of a structure, such as thebody of a worm, from the situation where itpasses out of the plane of section. Thesefactors make attempts to reconstruct mor-phology from such ‘natural’ splits difficultand feasible only when a large number ofspecimens are known or when the mor-phology is relatively simple. If essentialmorphological data are to be obtained forthe majority of taxa, to which neither ofthese conditions apply, then as much ofthe three-dimensional form as possiblemust be extracted from a single specimen.

The simplest and most commonapproach to determining the three-dimen-sional form of a fossil is to separate it fromthe matrix, either chemically or mechani-cally. Chemical dissolution techniques,though widely used in many branches ofpalaeontology (including the recovery ofKonservat-Lagerstätte fossils such as theCambrian ‘Orsten’ faunas; see Müller etal. 1995) require fossils and matrix to bechemically dissimilar. Although dissolutionhas been used to extract microfossils suchas radiolarians from the Herefordshirenodules (Briggs et al. 1996), the predomi-nantly calcitic nature of both matrix andmacrofossils precludes its use in the studyof most of the faunal elements. Mechani-cal excavation of the Herefordshire fossilsis possible and has some utility for thestudy of large and robust structures (e.g.,Briggs et al. 1996, fig. 1c, e). However, it isboth extremely time consuming and inap-propriate for resolving small or fragilestructures, and thus is not a satisfactoryapproach for the study of the fauna as awhole. In view of these difficulties, ourstudy has concentrated not on the physi-cal isolation of the fossils but on tech-niques based on serial slices, in which

many cross-sections are obtained fromparallel planes intersecting the fossil. Dataof this form can be collected in several dif-ferent ways, but all approaches producedatasets that can be reconstructed intothree dimensions with the aid of a com-puter.

APPROACHES TO SLICE DATA ACQUISITION

Non-destructive approaches

Two commonly used medical scan-ning technologies, Computed Tomography(CT) and Magnetic Resonance Imaging(MRI), use non-invasive techniques to pro-duce ‘slice-based’ datasets representingrespectively, the X-ray absorption, anddegree of magnetically induced nuclearresonance of materials in the sample. Thelatter technique is not well suited to solidgeological materials, but CT has beenused successfully to provide three-dimen-sional data on fossils (Hamada et al. 1991;Torres 1999). Although our preliminaryexperiments have shown that the Here-fordshire fossils within nodules can bedetected by CT scanners, generally avail-able equipment is capable at best of slicespacings and pixel sizes of around 0.3 mmto 0.5 mm. Specialised CT equipment(e.g., the University of Texas High Resolu-tion X-ray Computed Tomography Facility)can, under ideal conditions, achieve reso-lutions over an order of magnitude finerthan this, and might thus in principle be aviable tool for studying the fauna. How-ever, time on such equipment is expensiveand its ability to consistently distinguishfossil from matrix in the Herefordshirematerial at high resolutions is as yetunproven. While CT may provide a valu-able tool for investigating large orextremely rare taxa, we do not consider itto be a viable approach to the routinestudy of this fauna.

3


Serial slicing

Serial sectioning of fossils is the nor-mal approach to the collection of ‘slice-based’ datasets. However, the term ‘serialsectioning’ is commonly used to encom-pass both serial slicing by saw or bladeand serial grinding (Ager 1965, p. 213).These are substantially different tech-niques, and we prefer to use the separateterms to distinguish between them.

Serial slicing is less commonly usedon fossils than serial grinding, although itdoes allow much of the original material tobe retained whilst providing the requiredmultiplanar views. The method is bestsuited to larger specimens than the Here-fordshire fossils because intact slices sub-stantially thinner than 1 mm are extremelydifficult to produce by sawing. In contrast,biological microtome equipment can gen-erate exceptionally fine slices (as small as1 µm). However, microtomy requires anon-brittle sample capable of forming athin cohesive slice—properties not associ-ated with crystalline materials. Almost cer-tainly for these reasons, experiments onthe use of microtomy on Herefordshirefossils were unsuccessful, despite theirvacuum impregnation with resin. Serial grinding

Serial grinding involves the sequentialremoval by abrasion of small thicknessesof material from a single planar surfacewhich is photographed at each stage.Although this approach is destructive, itsuffers from none of the problems of a lackof cohesiveness associated with slicing,and inherently produces very flat polishedsurfaces which are ideal for photography.For these reasons we consider it the onlyviable approach to obtaining high-qualityserial images from the Herefordshire fos-sils. Serial grinding is not a new techniqueand has been used extensively on fossilsof widely differing sizes (e.g., Sollas andSollas 1913; Stensiø 1927; Ager 1956-67;

Baker 1978; Hammer 1999). Grinding hasbeen carried out by other researchers witha variety of pieces of dedicated equip-ment, the most popular of which havebeen variants of the ‘Croft Grinder’ (Croft1950; see also Ager 1965; Sandy 1989).The principle differences between our dataacquisition technique and those used bymost previous authors are the relative sim-plicity of the grinding device, and themethod of data capture—photography atregular intervals rather than manually pre-pared drawings often at irregular intervals.Considerations for Data Acquisition by Serial Grinding Photography.

The destructive nature of serial grind-ing places special demands on the pho-tography of each freshly exposed surface.The photographic technique used mustnot only capture as much information aspossible, but it must have a very low fail-ure rate, because images cannot be re-taken after grinding has progressed. Digi-tal rather than conventional photography istherefore a key element of our methodsince digital images can be downloadedinstantly to a computer where they can bechecked for quality before grinding contin-ues. The use of a digital camera alsooffers significant economic savings as wellas removing the need for images to bescanned into a computer for reconstruc-tion.

Registration. Although the process-ing of images from serial grinding is dis-cussed below, one aspect must beconsidered here as it impacts on the earlystages of the technique. Before any recon-struction can be attempted images mustbe ‘registered’ (or aligned) so that thearrays of pixels line up correctly. This canbe accomplished in one of two ways: 1)the sample can be placed in exactly thesame position relative to the field of viewof the camera for each image capture; or2) the sample can be processed so that it

4


contains physical ‘fiduciary’ (or alignment)markings of some sort, that can be usedsubsequently to align the images. Thesecond approach is simpler in practiceand is preferred in our work. Fiduciarymarkings used in similar work by previousauthors are, typically, straight holes drilledperpendicular to the grinding surface sothat they appear in the corners of theimage (e.g., Herbert 1999, figure 3.2;Baker 1978). However, holes less than 0.5mm in diameter cannot be produced reli-ably with current drilling technology andlarger holes would be unacceptable in thecontext of this study. Consequently, weuse two perpendicular edges (see below)as fiduciary structures, which are roughlyaligned in the field of view when imagesare first captured. These edges are usedto guide precise digital alignment duringthe registration phase of post-processing(see Appendix 1, section 1) .

METHODOLOGY FOR GRINDING AND DATA CAPTURE

Specimens selected for serial grinding(Figure 2.1) are first sawed to produce asmall block with two faces perpendicularto each other and to the direction of grind-ing, and as close to the specimen as pos-sible. These side faces are then polishedto obtain the sharp edges at their junctionwith the ground face (Figure 2.2). Theblock is set within a resin cuboid or cylin-der (Figure 2.3), that is then mounted withwax onto a Wirtz Buehler Slide Holder(Figure 2.4). Note that while Figure 2 illus-trates a grinding run performed on a singlefossil, multiple specimens are normallymounted within one resin block for greaterefficiency. The slide-holder consists of aninner metal cylinder that, by means of ascrew thread, can be positioned accu-rately to protrude any required distancebeyond the plane of an outer hard ceramicdisc. This enables a fixed thickness to be

ground away before the ceramic comesinto contact with the grinding plate andprevents any further removal of material.

Specimens are ground manually on aglass plate (Figure 2.5) with 600 gradecarborundum powder and water to removethicknesses of approximately 30-50 µm ata time. After each thickness is removed,the slide holder is inverted and placed onits base. The surface is then washed andpolished for a few seconds with a small flatplate coated in fine-grade aluminium oxidepowder and water (Figure 2.6). This finalpolishing improves image quality withoutremoving any appreciable thickness ofsample. The surface is then washedagain, and the slide-holder placed under aLeica MZ8 binocular microscope. Animage of each specimen and alignmentedges (Figure 2.7) is then captured bymeans of a digital camera attached to themicroscope operating at 1000 by 800 pixelresolution in 24-bit colour. The image ischecked for quality prior to filing as part of

Figure 2. Grinding process.

5


a sequential set. All images for a grindingrun are taken at the same magnification,with the same camera settings, and underconsistent lighting conditions; only thefocus is re-adjusted for each image. Thegrinding process is then repeated for asmany iterations as required. The timetaken per image depends on how manyspecimens are mounted at once for agrinding run, but on average it takes sixminutes to grind a 30 µm interval and oneminute to photograph each specimen.After completion of a grinding run eachdataset is prepared for use by registrationof the images (see Appendix 1, section 1).

VISUALIZATION AND RECONSTRUCTION

Slice videos

Computer video files can be assem-bled directly from the datasets resultingfrom grinding (see Appendix 1, section 2),providing a simple way of leafing back-wards and forwards through the sliceimages (Figure 3). These files form aninvaluable tool for the exploration ofdatasets and, although less striking thanfull three-dimensional reconstructions, areoften more useful for tracing the originsand termination of small or obscure struc-tures. Nonetheless, they do not constitute

a three-dimensional reconstruction in theaccepted sense and are best used to com-plement rather than substitute for suchreconstructions. Construction of three-dimensional models

The ‘surfaces’ method. Two method-ologies exist for the construction of three-dimensional models from data based onslices (Brown and Herbert 1996; Herbert1999). The first, the ‘surfaces’ approach(Herbert 1999; see also Chapman 1989),begins with a manual interpretation step,tracing all structures of interest from eachslice image (Figure 4.1-4.3). Traced out-lines are digitized into line segments form-ing closed loops (Figure 4.4). Thereconstruction algorithm assembles thesestacked two-dimensional loops into athree-dimensional geometry, typically amesh of triangular facets (Figure 4.5),which can be rendered into a two-dimen-sional view by a number of standard com-puter graphics techniques. The two keyadvantages of this method are that it, (1)does not require photographic images of

Figure 3. Slice video showing raw dataset (distal por-tion of monograptid graptolite; OUM C.29526a).

Figure 4. Vector-based reconstruction process of Her-bert (1999).

6


the slices and, (2) is capable of workingwith low-quality, sparsely-spaced data.Neither of these is relevant to the presentstudy. This method also has several disad-vantages, especially in the context of theHerefordshire fauna. The initial interpreta-tion stage is extremely time-consumingwhen a large number of slice images areinvolved, and also threatens the objectivityof the reconstruction as structures must bedelineated before the three-dimensionalform becomes apparent. When the mar-gins of the fossil structures are unclear (asis often the case with the Herefordshirefossils), there is a danger that subjectiveinterpretations will influence the recon-structed morphology. In addition, the algo-rithmic reconstruction of stacked two-dimensional loops into three-dimensionalsurfaces is subject to the ‘correspondenceproblem’–that is, the difficulty of determin-ing the manner in which loops in subse-quent slices connect to each other.Herbert (1999) approached this problemwith algorithms that made a priori assump-tions about the morphology of the object tobe reconstructed, an approach that isclearly not viable for the reconstruction ofpoorly understood morphologies such asthose of many Herefordshire fossils.Other available approaches require theuser to determine correspondence, a pro-cedure that is both time consuming andrepresents another potential source ofsubjectivity.

Volume-based methods. The secondreconstruction methodology, that of vol-ume-based reconstruction, is preferred inthe present study. Volume representationis a three-dimensional extension of ‘raster’or ‘bitmap’ representations of two-dimen-sional images, in which an image is mod-elled as a rectangular array of regularlyspaced pixels. A volume dataset thus con-sists of a three-dimensional array of regu-larly spaced volume elements or ‘voxels’,each stored as a number representing the

value of some property of the object at thatpoint. The sequential slice images gener-ated by the grinding technique (after minorpost-processing — see Appendix 1, sec-tion 3), represent a volume of this sort inwhich the values of the voxels representthe reflectivity of the surface as capturedby the digital camera. Volume-basedthree-dimensional reconstructions, largelyutilizing datasets from MR or CT scanners,are routinely used in medical applicationsand have been used to produce volume-based reconstructions of fossils (e.g.,Hamada et al. 1991; Torres 1999). Thesemethods have also been applied to opti-cally captured volumes. Brown and Her-bert (1996) attempted to produce volume-based models from serial section data,although the low resolution of the grapto-lite sections on which that work was basedprecluded a successful reconstruction.More successfully, Hammer (1999) recon-structed a halysitid coral colony from vol-ume data obtained by serial grinding andcaptured with a flatbed scanner.

Volume rendering. There are varioustechniques, collectively known as volumerendering, for the direct visualization ofvolume data . Although they vary in detail,all involve forms of ‘ray casting’, where thecomputer calculates an array of virtuallight rays ‘shone’ into the volume from aspecified angle, determines their fateaccording to mathematical rules based onthe values of voxels in the volume, andbuilds a two-dimensional image from theresults of these calculations. We haveused mean-intensity volume rendering,where the final intensity of the virtual ray isdetermined by the mean values of the vox-els it encounters. This algorithm producesa virtual ‘X-ray’ image of the volume fromany angle the user chooses (Figure 5).Like true X-radiographs, the images are arelatively poor aid to visualizing the outersurface of objects, but are effective forimaging of large or strongly absorbent

7


(=dark) structures that otherwise may beobscured by less substantial features.

Isosurfaces. While volume data canthus be rendered directly, the preferredapproach for medical visualization fromCT or MR data is usually to generate poly-gon mesh surfaces known as isosurfaces.These are generated by algorithms thatconnect points of a constant (user-deter-mined) intensity within the volume. If thethreshold level is chosen correctly, an iso-surface generated from a dataset such asshown in Figure 3 should correspond tothe original external surface of the organ-ism. Isosurfaces can be rendered withany of several computer graphics tech-niques available for the visualization ofgeometrically defined surfaces. Ray-trac-ing, though the most computationallyexpensive of these techniques, providesthe clearest and most natural visualiza-tions (Figure 6). Although volume-ren-dered ‘X-ray’ reconstructions have someapplication to the Herefordshire fossils,ray-traced isosurfaces are our preferredmeans of three-dimensional visualization.Reconstructions produced in this wayretain the objectivity of volume data with-out compromising the clarity and sharp-ness of the rendered images.

Animation. Isosurfaces have beenused previously to produce static imagesof fossils (e.g., Torres 1999; Hammer1999; Brown and Herbert 1996). Whilesuch images have some illustrative value,a true appreciation of three-dimensionalmorphology is best achieved with an inter-active model. We produce these (for bothisosurface and volume rendering tech-niques) from a batch of images renderedfrom sequential angles and assembledinto a computer video file that can be pre-sented as a reconstruction of the speci-men (Figures 7-11). As with video filesassembled directly from slice images,these files can be played backwards orforwards and provide a simple but effec-tive means of viewing the reconstructionfrom a variety of angles. They also enablethe reconstruction to be distributed easilyfor viewing on any modern computer with-out the need for dedicated software.Pitfalls for three-dimensional reconstructions

Transparency of calcite. The natureof the Herefordshire material necessitatesguarding against two potential pitfalls. Thefirst concerns the assumption that eachimage represents the intersection of thefossil with a single plane. The calcite thatcomprises the fossils is normally dark and

Figure 5. Volume-rendered ‘X-ray’ visualization of distalportion of monograptid graptolite; OUM C.29526a. 165slices at 30 µm interval. Specimen is approximately 5 mmlong.

Figure 6. Ray-traced isosurface visualization of anteriorportion of a cranidium of the trilobite Tapinocalymenevulpecula Siveter 1980; OUM C.29527a. 172 slices at50 µm interval. Specimen is approximately 16 mmwide.Top is anterior.

8


opaque, but translucent regions canreveal features beneath the surface planeor be relatively light in shade when matrixlies only a short distance below the plane.Such occurrences, which have the poten-tial to invalidate reconstructions partially,are detected by inspection of the sliceimages and corrected by manual retouch-ing (see Appendix 1, Section 3).

‘Halo’ effects. Light-coloured regions,presumed to be reduction haloes, often

occur around specimens of certain taxa(Figure 12). Fossil material within thesehaloes is often no darker than the matrixoutside them, making the selection of asingle isosurface threshold level for thewhole fossil impossible. Reconstructionsmay therefore fail to include extremitieswithin these regions that are nonethelessclearly visible on the slice images. Whilethis problem can be overcome by carefulmanual darkening of these regions, a sim-pler alternative is often to prepare a dis-section (see below) of the inside of a light

Figure 7. AVI forat video file of ray-traced isosurfacereconstruction of distal portion of monograptid graptolite;OUM C.29526a. 165 slices at 30 µm interval. Specimenis approximately 5 mm long.

Figure 8. AVI format video file of Volume-rendered ‘X-ray’reconstruction of distal portion of monograptid graptolite;OUM C.29526a. 165 slices at 30 µm interval. Specimen isapproximately 5 mm long.

Figure 9. AVI format video file of ray-traced isosurfacereconstruction of anterior portion of a cranidium of thetrilobite Tapinocalymene vulpecula Siveter 1980; OUMC.29527a. 172 slices at 50 µm interval. Specimen isapproximately 16 mm wide.

Figure 10. AVI format video file of ray-traced isosurfacereconstruction of anterior fragment of the mollusc Acae-noplax hayae Sutton et al. 2001; OUM C.29532. 190slices at 30 µm interval. Specimen is approximately 5mm long.

9


region, and to reconstruct this section sep-arately.

EXTENSIONS OF RECONSTRUCTION TECHNIQUES

Whole fossil reconstructions

Reconstructions from a single grindingrun typically comprise just part of a fossil,as grinding normally commences from theplane along which the nodule was origi-nally fractured. Reconstructions of entire

fossils may be obtained by gluing part andcounterpart together before sawing to sizeso that the entire fossil can then be groundthrough as one unit. However, thisapproach can produce artifacts fromimperfect alignment of the glued surfacesor from the glue itself. Alternatively andmore satisfactorily, part and counterpartcan be reconstructed separately and reas-sembled digitally before rendering. Figure13 is an example of a whole fossil recon-structed in this way. Dissection of models

Reconstructions of portions of a fossil,or of a subset of the structures present,are often of great utility. A strength of thevolume-based approach described here isthe ease with which such ‘dissected’reconstructions can be prepared. Copiesof the appropriate slice images are simplymanipulated using image editing softwareto remove structures or portions of the fos-sil that are not required (see Appendix 1,section 3), and are then rendered in thenormal way. Slice videos and pre-dissec-tion three-dimensional reconstructions

Figure 11. AVI format video file of ray-traced isosurfacereconstruction of antero-median fragment of the mol-lusc Acaenoplax hayae Sutton et al. 2001; OUMC.29529. 243 slices at 30 µm interval. Specimen isapproximately 7 mm long.

Figure 12. Reduction haloes surrounding an unnamed polychaete annelid; OUM C.29525a.

10


provide the ideal tools to guide such dis-section work, and ensure that unwantedstructures are removed accurately.

APPLICATIONS OF RECONSTRUCTION TECHNIQUES

The techniques described herein,although largely illustrated with well-understood biomineralized taxa, havebeen developed to elucidate the morphol-ogy of the essentially soft-bodied organ-isms that dominate the Herefordshirefauna (such as Acaenoplax hayae, Fig-ures 10-11, see also Sutton et al. 2001),and provide the primary means for palae-ontological studies. However, they areeasily adapted to other situations andessentially require only that the fossil pre-serves three-dimensional form and is visu-ally distinguishable from matrix. Despitethe destructive nature of grinding we thusexpect that similar applications, whosemain goal is simply to determine morphol-ogy, will arise in other situations where iso-lation of the fossils in question isimpractical.

These techniques also have broaderpotential applications within palaeontol-

ogy. Virtual models of fossils are usefultools for the dissemination of morphologi-cal information, as they contain more infor-mation than static images, yet can becopied, distributed, and viewed quicklyand at minimal cost. Applications in edu-cation can be envisaged as a means offamiliarizing students with morphology offossils, particularly where material is toolarge, small, or rare to be physicallymanipulated. Computer-based modelsmay also prove to be an efficient way ofconveying morphology to professionalpalaeontologists, particularly those lessfamiliar with the group in question. Thesetechniques might also provide a methodol-ogy for capturing morphological informa-tion as a prelude to further computer-based modelling work. Such studies mightinclude the reconstruction of skeletalstructures (e.g., conodont feeding appara-tus) from isolated material, the illustrationof theoretical models of unpreserved softparts (e.g., brachiopod musculature), andstudies of functional morphology thatrequire digital models (e.g., finite elementstress analysis).

ACKNOWLEDGEMENTS

We thank Jenny Corrigan, Juliet Hay,Jeremy Hyde, and Philip Jackson forassistance with various aspects of speci-men processing. This work is funded byNERC Grant GR3/12053.

REFERENCES

Allison, P. A., and Briggs, D. E. G. 1991. p. 120-140. Thetaphonomy of soft-bodied animals. In Donovan, S. K.(ed.), Fossilization: the process of taphonomy.Belhaven Press, London.

Ager, D. V. 1956-67. A Monograph of the British LiassicRhynchonellidae. Palaeontographical Society,Monographs, p. 1-172.

Ager, D. V. 1965. Serial grinding techniques. p. 212-224.In Kummel, B. and Raup, D. (eds.), Handbook ofPaleontological Techniques. W. H. Freeman, NewYork.

Figure 13. AVI format video file of ray-traced isosur-face reconstruction integrating part and counterpart ofa cranidium of the trilobite Tapinocalymene vulpeculaSiveter 1980; OUM C.29527a-b. 29527a (anterior): 172slices at 50¨µm interval. 29527b (posterior): 197 slicesat 50 µm interval. Specimen is approximately 13.5 mmlong medially.

11


Baker, P. G. 1978. A technique for the accurate recon-struction of internal structures of micromorphic fos-sils. Palaeontology, 19:565-584.

Briggs, D. E. G., Siveter, David J., and Siveter, Derek J.1996. Soft-bodied fossils from a Silurian volcaniclas-tic deposit. Nature, 382:248-250.

Brown, I. M., and Herbert, M. J. 1996. Knowledge-basedgeological visualisation using AVS. Electronic Geol-ogy [online], 1 (Special volume on the proceedingsof the Geoscience Information Group on Geologicalvisualisation; the intelligent picture?):37-47. Availablefrom Internet: http://www.electronicjournals.co.uk/eg/pdf/EG1-5.pdf

Chapman, R. E. 1989. Computer assembly of serial sec-tions. p. 157-164. In Feldmann, R. M., Chapman, R.E., and Hannibal, J. T. (eds.), Paleotechniques.Paleontological Society Special Publication 4.

Croft, W. N. 1950. A parallel grinding instrument for theinvestigation of fossils by serial sections. Journal ofPaleontology, 24:693-698.

Hamada, T., Tateno, S., and Suzuki, N. 1991. Threedimensional reconstruction of fossils with X-ray andcomputer graphics. Scientific Papers of the Col-lege of Arts and Sciences, University of Tokyo,41:107-118.

Hammer, Ø. Computer-aided study of growth patterns intabulate corals, exemplified by Catenipora heintzifrom Ringerike, Oslo Region. Norsk GeologiskTidsskrift, 79:219-226.

Herbert, M. J. 1999. Computer-based Serial SectionReconstruction. p. 93-126. In Harper, D. A. T. (ed.),Numerical Palaeobiology: Computer-based mod-

elling and analysis of fossils and their distribu-tions. John Wiley and Sons, Chichester.

Müller, K. J., Walossek, D., and Zakharov, A. 1995.‘Orsten’ type phosphatized soft-integument preserva-tion and a new record from the Middle CambrianKuonamka Formation in Siberia. Neues Jahrbuchfür Geologie und Paläontologie Abhandludgen,197:101-118.

Orr, P. J., Briggs, D. E. G., Siveter, David J., and Siveter,Derek J. 2000. Three-dimensional preservation of anon-biomineralised arthropod in concretions in Sil-urian volcaniclastic rocks from Herefordshire,England. Journal of the Geological Society, Lon-don, 157:173-186.

Sandy, M. R. 1989. Preparation of serial sections. p.143-150. In Feldmann, R. M., Chapman, R. E., andHannibal, J. T. (eds.), Paleotechniques. Paleonto-logical Society Special Publication 4.

Sollas, I. B. J., and Sollas, W. J. 1913. A study of theskull of a Dicynodon by means of serial sections.Philosophical Transactions of the Royal Societyof London, 204: 201-225.

Stensio, E. A. 1927. The Downtonian and Devonian ver-tebrates of Spitzbergen. Skrift, Svalbard Nordis-havet, 12: 1-31.

Sutton, M. D., Briggs, D. E. G., Siveter, David J., andSiveter, Derek J. An exceptionally preserved vermi-form mollusc from the Silurian of England. Nature,410: 461-463.

Torres, A. M. 1999. A three-dimensional CT (CAT) scanthrough a rock with Permian alga Ivanovia tebagaen-sis. Journal of Paleontology, 73:154-158.

12


APPENDIX 1: DETAILS OF COMPUTER-BASED PROCESSING

This represents a step-by-step docu-mentation of the computer-based ele-ments of the reconstruction process usedto produce slice videos and isosurfacemodels. For each step we provide both a‘package-independent’ overview (in bold),followed by ‘package-specific’ details ofour method. The latter are intended onlyas a description of our procedures, not asa set of instructions for those unfamiliarwith the software involved, nor indeed asan exclusive recommendation of any spe-cific software package—equivalents of allmajor packages used are available formost platforms. Details of software pack-ages used in the study are given inAppendix 2, and terminology specific tothese packages is placed within invertedcommas on first usage.1) Registration and cropping

Images are first registered(aligned), using the alignment edges inthe images as a guide. An image-editing software package is used toshift and rotate the images manually,so that the edges are in exactly thesame position. Once registered, theimages are cropped to remove extrane-ous background, including registrationedges. All images must be croppedusing exactly the same parameters sothat alignment is preserved.

One image, normally the first, is cho-sen as the datum for alignment. Thisimage is loaded into PhotoPaint, and anew ‘object’ is created to lie above thebackground. Straight lines of a single pixelin width are drawn on this object to corre-spond to the alignment edges of the imagebeneath. The ‘Eraser’ tool is then used toremove portions of these lines, so theyappear as coarse dashes. Oblique linesare also drawn on the object from edge toedge near each of the four corners of the

image; these ensure that the object isexactly the same size as the image, whichassists with placement in the next step.Once the alignment object is complete it isplaced into the clipboard with a copy com-mand and the file closed.

The registration process now com-mences, and the following procedure isperformed for each image. The image tobe aligned is loaded, and the alignmentobject is inserted with the ‘Paste As NewObject’ command—this alignment objectis not moved subsequently, but used as aguide for aligning the image. The imageitself is converted to an object with the‘Create From Background’ command. Thisimage object is then shifted manually withthe ‘Object Picker’ tool selected, by use ofthe cursor keys, so that the point wherethe two alignment edges meet is correctlypositioned. The rotational centre of theobject is then dragged to that point, andthe correct rotation of the image applied bymeans of trial and error entries of anglesin the ‘Tool Settings Roll-up’. When andonly when the correct rotation is deter-mined, it is applied with the ‘Transform’button on the roll-up, and any final single-pixel shifts required for perfect alignmentare performed with the cursor keys.Finally, the alignment object is removedfrom the image which is saved in its origi-nal format. This process must be carriedout for each image in the dataset and isthus fairly time-consuming. Much of thework can be automated, however, byrecording actions as ‘scripts’. We normallymake use of one script to create the nec-essary objects, and another to remove thealignment object after alignment. In prac-tice, we find that it is possible to registerup to 60 images per hour.

After registration, the images arecropped to size with the ‘Deskew Crop’tool. Care must be taken that the crop

13


area set includes all of the fossil on everyimage. Once the correct crop size is deter-mined, the action is recorded as a script,and then applied to all images with the‘Batch Playback’ command.2) Post-processing for preparation of slice videos

Slice videos, if required, can be pre-pared directly from the registered andcropped image set using a suitablesoftware package. Typically such vid-eos are prepared from a second copyof this dataset, the images of which arecolour balanced and/or reduced in sizeprior to video production. This pro-cessing is carried out with image-edit-ing software. Size reduction isnecessary only if the resulting videowould otherwise be too large to storeor play on any given computer.

To prepare for video production, acopy of the registered and cropped imageset is made, and the following PhotoPaintoperations are applied first to one image,recorded as a script, and then applied toeach image by means of the Batch Play-back command.

a. Colour is balanced to taste with the‘Level Equalization’ command.

b. Image size is reduced, if necessary,with the ‘Resample’ command.

Processed images are opened as afile sequence video in Ray Dream, andthen saved as an AVI video file in Cinepakformat.3) Post-processing for isosurface reconstruction

To produce a dataset suitable forthe isosurface algorithm, the registeredand cropped dataset must be con-verted from colour to monochrome,inverted (so that fossil is light andmatrix dark), and reduced in size if therestrictions of the isosurface packageso demand. We find that the Hereford-

shire fossils are usually more clearlydistinguished from the matrix if theconversion to monochrome is per-formed using the red channel only.These operations, which must be iden-tically performed on all images in thedataset, are carried out with an imageediting software package.

Any required retouching of imagesshould be carried out at this step,although in most cases we prefer toreconstruct an un-retouched or mini-mally retouched set of images first,then return to this point in the processto perform any required ‘cleaning’ on asecond copy of the dataset. Retouch-ing is carried out manually using animage-editing software package. Inmost cases it comprises the deletion ofobjects in the matrix that appear in thereconstruction but are clearly not partof the main fossil, or of structures visi-ble through translucent regions of cal-cite. Retouching can also include thebrightening of structures that areclearly part of the main fossil, but aretoo dark to appear in the reconstruc-tion (i.e., are falling below the chosenisosurface threshold, usually becauseof transparency problems). We find ithelpful to choose an isosurface thresh-old during the retouching step, so thatthese problems can be assessed withinthe image editing software package.

Post-processing operations are per-formed on each image using PhotoPaint.As in Step 2 above, they are first per-formed on one image, the actionsrecorded as a script, and then applied toeach image by means of the Batch Play-back command.

a. The image is converted to mono-chrome. This can be accomplished withthe ‘Convert to Greyscale (8-bit)’ com-mand. If only the red channel is required(see above), the Level Equalization com-mand is used to remove the green and

14


blue signals, the Convert to Greyscale (8-bit) command is then applied, and finallythe ‘Brightness-Contrast-Intensity’ com-mand is used to enhance the visual clarityof the image (which will be too dark).

b. The image is inverted using the‘Invert’ command.

c. If necessary, the image is resam-pled to a smaller size. AVS (see below)can only accept volumes up to 255 x 255 x255 voxels in size, so the image size mustbe reduced to 255 x 255 pixels or smallerbefore a reconstruction can be attempted.Resampling ideally is carried out with anintegral divisor (e.g., at 50% or 33.33%);arbitrary values can introduce an elementof blurring, although this is rarely of criticalimportance. The final image width mustalso be a multiple of 4 to satisfyAVSCONV. This can be accomplishedeither by padding with black space usingthe ‘Paper Size’ command, or by furthercareful cropping.

Retouching, if required, is also carriedout in PhotoPaint, mainly by use of the‘Eraser’ tool to delete objects, and the‘Effect’ tool to brighten others. The‘Threshold’ command can be used toapply the chosen isosurface thresholdlevel and determine what will and what willnot appear in the reconstruction. Thethreshold effect can then simply beremoved with the ‘Undo’ command.4) Isosurface production

The post-processed image set mustfirst be converted into whatever volumeformat the isosurface softwarerequires, using appropriate software.Once informed of the appropriatethreshold level the isosurface softwarecan then generate the surface. The sur-face may then need to be converted toa format that can be read by the ray-tracing software employed.

We convert the dataset to AVS volumeformat using the custom AVSCONV pro-

gram. This requires the images be in Win-dows BMP format (uncompressed), beconsecutively numbered with three digits(i.e., file names of the formName001.bmp, Name002.bmp), andplaced in a single directory. The userselects the first file, clicks the process but-ton, and when requested enters the num-ber of files in the dataset. On completion,the program places the AVS format datafile in the same directory, with the name‘output.dat’.

An AVS ‘network’ is created, as fol-lows. The data file produced in Step 3 isread with the ‘read volume’ module, whichfeeds the ‘average down’ module. Thisshould be set with relatively high x, y, andz coefficients (2 or more) at this step. Theoutput from the ‘average down’ module isfed to the right hand input port of the ‘isos-urface’ module, the output of which is sentto the right hand port of the ‘geometryviewer’ module. The isosurface levelshould now be set to produce the bestpossible reconstruction; as noted abovewe normally find it helpful to choose thislevel during retouching. When the isosur-face appears correct, the coefficients ofthe ‘average down’ module should bereduced to 1, and the output of the ‘isosur-face’ module connected to the ‘writegeom’ module, which should be set to pro-duce a ‘DXF’ file. As the processinginvolved in the network can be time con-suming, we find it helpful to disable theflow executive whilst making changes toparameters of these modules.

DXF files produced by the ‘write geom’module can be very large. We find thatfiles much over 100MB in size are imprac-tical to render in Step 5 below. In somecases therefore, particularly with datasetscontaining a large number of images, thefinal isosurface is produced with x and ysettings of 2 rather than 1 in the ‘averagedown’ module. Although we normally pre-fer not to average down in the z axis, if

15


alignment errors are producing strong arti-facts (lineation parallel with the plane ofgrinding), a z setting of 2 can help tosmooth these out, as well as to reduceoutput file size.5) Rendering

Geometry files describing isosur-faces are read into a ray-tracing soft-ware package, or any other three-dimensional visualization package. Iso-surfaces are stretched to correct fordifferences in voxel spacing withinslices and between slices (this may bedone during Step 4 instead, if the soft-ware allows). If multiple isosurfaces areto be joined (e.g., Figure 11), they areboth imported into one scene, and thenmanually moved into a contiguousposition. The visualization package isthen used to generate a video file of thespecimen rotating in the desired fash-ion.

DXF files produced by AVS are readinto Ray Dream with the ‘Import’ com-mand. All the default import options areleft unaltered, except ‘Scaling ConversionFactor’ which is set to ‘0.5 cm’ rather than

‘1 in.’. Once imported, the isosurface isstretched in the z axis to correct for slicespacing—the z scale property of the isos-urface object is set to 100% x (Pixels/mmin resampled images) / (slices/mm). Anyother objects are imported and treated inthe same way, and carefully moved intoposition relative to each other. Whenbringing two objects together, we preferfirst to bring a single corresponding pointinto contact, move the ‘hotpoint of rotation’to that point, and then rotate the objectinto position using the ‘virtual trackball’. A‘shader’ is then applied to the object(s),lights placed in the scene, and an anima-tion set up to rotate the object in thedesired fashion over a time period of sev-eral seconds. The ‘production frame’ iscarefully positioned and sized to the mini-mum size necessary to contain the objectat all steps of the animation. The anima-tion is then rendered within Ray Dream,normally with an image width of 300 to400pixels and a frame rate of 15 or 18 framesper second. It is saved as an AVI video filein Cinepak format, which forms the finalreconstruction.

16


APPENDIX 2:SOFTWARE AND EQUIPMENT

1) Commercial Software

Image editing:PhotoPaint version 8 (for MicrosoftWindows)Corel CorporationCorporate Headquarters1600 Carling AvenueOttawa, OntarioK1Z 8R7, Canadawww.corel.comRendering and production of slice vid-eos:Ray Dream 3D version 1.02 (forMicrosoft Windows)MetaCreations Corporation6303 Carpinteria Ave.Carpinteria, CA 93013, USAwww.metacreations.comIsosurface production:AVS version 5 (for Solaris)Advanced Visual Systems Inc.300 Fifth Avenue Waltham, MA 0245, USAwww.avs.com

2) Free Software

Conversion of image files to AVS vol-ume format:

AVSCONV version 1.1 (forMicrosoft Windows)Author: Mark Sutton

URL for download: palaeo-eletronica/2001_1/s2/programExport from AVS geometry format toDXF for import into RayDream:‘write geom’ AVS module [Must beinstalled into AVS system]Author: Georg ThallingerURL for download: http://www.iavsc.org/repository/avs5/source/data_output/write_geom/

3) Equipment used

SV Micro Digital Microscope CameraSound Vision, Inc. 492 Old Connecticut PathFramingham, MA 01701http://www.soundvisioninc.com/Wirtz-Buehler ‘Slide Holder’ (hand-held grinder)Buehler Krautkramer UK Milburn Hill RoadUniversity of Warwick Science Park Coventry CV4 7HS United Kingdomhttp://www.buehlerltd.com/

17

METHODOLOGIES FOR THE VISUALIZATION AND RECONSTRUCTION OF THREE-DIMENSIONAL FOSSILS FROM THE SILU...Mark D. Sutton, Derek E. G. Briggs, David J. Siveter, and Derek J. SiveterABSTRACTINTRODUCTIONAPPROACHES TO SLICE DATA ACQUISITIONNon-destructive approachesSerial slicingSerial grindingConsiderations for Data Acquisition by Serial Grinding Photography.

METHODOLOGY FOR GRINDING AND DATA CAPTUREVISUALIZATION AND RECONSTRUCTIONSlice videosConstruction of three-dimensional modelsPitfalls for three-dimensional reconstructions

EXTENSIONS OF RECONSTRUCTION TECHNIQUESWhole fossil reconstructionsDissection of models

APPLICATIONS OF RECONSTRUCTION TECHNIQUESACKNOWLEDGEMENTSREFERENCESAPPENDIX 1: DETAILS OF COMPUTER-BASED PROCESSING1) Registration and cropping2) Post-processing for preparation of slice videos3) Post-processing for isosurface reconstruction4) Isosurface production5) Rendering

APPENDIX 2: SOFTWARE AND EQUIPMENT1) Commercial Software2) Free Software3) Equipment used

FIGURESFigure 1. Detail of fine spines preserved on margin of Acaenoplax hayae Sutton et al. (2001), fro...Figure 2. Grinding process.Figure 3. Slice video showing raw dataset (distal portion of monograptid graptolite; OUM C.29526a).Figure 4. Vector-based reconstruction process of Herbert (1999).Figure 5. Volume-rendered ‘X-ray’ visualization of distal portion of monograptid graptolite; OUM ...Figure 6. Ray-traced isosurface visualization of anterior portion of a cranidium of the trilobite...Figure 7. AVI forat video file of ray-traced isosurface reconstruction of distal portion of monog...Figure 8. AVI format video file of Volume-rendered ‘X-ray’ reconstruction of distal portion of mo...Figure 9. AVI format video file of ray-traced isosurface reconstruction of anterior portion of a ...Figure 10. AVI format video file of ray-traced isosurface reconstruction of anterior fragment of ...Figure 11. AVI format video file of ray-traced isosurface reconstruction of antero-median fragmen...Figure 12. Reduction haloes surrounding an unnamed polychaete annelid; OUM C.29525a.Figure 13. AVI format video file of ray-traced isosurface reconstruction integrating part and cou...

methodologies for the visualization ... - palaeo-electronica · ment, the most popular of which...

Documents