cultural heritage predictive rendering

DOI: 10.1111/j.1467-8659.2012.02098.x COMPUTER GRAPHICS forumVolume 0 (2012), number 0 pp. 1–14

Cultural Heritage Predictive Rendering

Jassim Happa1, Tom Bashford-Rogers1, Alexander Wilkie2, Alessandro Artusi3,4, Kurt Debattista1 and Alan Chalmers1

1International Digital Laboratory, University of Warwick, UK2Charles University in Prague, Czech Republic

3CASToRC Cyprus Institute, Cyprus4University of Girona UdG, Spain

[email protected]

AbstractHigh-fidelity rendering can be used to investigate Cultural Heritage (CH) sites in a scientifically rigorous manner.However, a high degree of realism in the reconstruction of a CH site can be misleading insofar as it can be seento imply a high degree of certainty about the displayed scene—which is frequently not the case, especially wheninvestigating the past. So far, little effort has gone into adapting and formulating a Predictive Rendering pipelinefor CH research applications. In this paper, we first discuss the goals and the workflow of CH reconstructions ingeneral, as well as those of traditional Predictive Rendering. Based on this, we then propose a research frameworkfor CH research, which we refer to as ‘Cultural Heritage Predictive Rendering’ (CHPR). This is an extension toPredictive Rendering that introduces a temporal component and addresses uncertainty that is important for thescene’s historical interpretation. To demonstrate these concepts, two example case studies are detailed.

Keywords: cultural heritage, virtual reconstruction, physically based rendering, global illumination, experimentalarchaeology, virtual archaeology, image-based lighting, High Dynamic Range imaging, sky modelling, flamemodelling, visual perception

ACM CCS: I.3.7[Computer Graphics]: Three-Dimensional Graphics and Realism; I.3.8 [Computer Graphics]:Applications.

1. Introduction

High-fidelity rendering allows for the investigation of Cul-tural Heritage (CH) using physically based methods. ThePredictive Rendering pipeline [GTS*97] presented the keyprinciples to render and validate real-world scenes. In juxta-posing the image synthesis pipeline for modern scenes withhistorical sites, however, it becomes apparent that in its cur-rent state, it only satisfies renditions of sites as they standtoday. For historical representations, we must identify andadd appropriate, accurate Scene Metadata, including its tem-poral components. Scene Metadata here refer to informationthat is meaningful in its interpretation for the historical expertobserver. This includes scene setup (geometry, materials andillumination) based on how people and natural forces interactwith the site; incorporating both physically and historicallybased properties to deliver plausible images.

We propose a novel research framework to reverse-engineer past sites by extending Predictive Rendering tomake use of extant objects and records, experimental dataand expert opinion. Through Cultural Heritage PredictiveRendering (CHPR), the aim is to investigate and understandthe motivation, function and visual perception of CH sitesin the present and past. Two case studies have aided in itsdevelopment: Panagia Angeloktisti, a Byzantine church onCyprus and the Red Monastery, a Coptic site in Egypt.

1.1. Problems in recreating the past

Efforts are being made to standardize methods for virtualCH research [Gro96, DCM95, 3dC09, AG07, DH09, GB11].However, achieving reliable rendering results for CH re-search remains a demanding challenge for several reasons,including:

c© 2012 The AuthorsComputer Graphics Forum c© 2012 The EurographicsAssociation and Blackwell Publishing Ltd. Published byBlackwell Publishing, 9600 Garsington Road, Oxford OX42DQ, UK and 350 Main Street, Malden, MA 02148, USA. 1

2 J. Happa et al. / Cultural Heritage Predictive Rendering

(1) Few rendering standards have been established andwidely adopted specifically for CH rendition.

(2) Validation of the past: physical condition and accessto historical records can lead to much interpretation.

(3) Sensitivity-of-CH concerns regarding data collection.(4) Current hardware and software limitations.

There are currently no universally accepted standards toassess accuracy of CH renditions. Metrics that partially con-tribute to documentation and validation are borrowed fromother sciences and disciplines. Examples include (but arenot limited to); Spectroradiometry, Laser Scanning [BR02],Ground Penetrating Radar (GPR) [CL10] and ReflectanceTransformation Imaging (RTI) [MGW01, FBM*06].

2. Background

2.1. Motivations for CHPR

Hypotheses about the past can be simulated to provide aresearch context in which a virtual rendition closely resem-bles its (known) physical counterpart. Currently, no generalframework for this exists, despite the many high-fidelitycase-study reconstructions that have been published in thelast 20+ years [HMD*10]. Nevertheless, several standardsexist related to documentation and data management ofCH. The CIDOC Conceptual Reference Model [Gro96],Dublin Core [DCM95], Functional Requirements for Biblio-graphic Records (FRBR) [IFL09], London Charter [DH09]and Sevilla Charter [GB11] are some examples. CHPR aimsnot to replace or compete with them, as it is not a frameworkfor data management or documentation purposes. Principlesfrom one or more existing standards can be used within aCHPR context. Our framework simply combines efforts dis-cussed in prior CH graphics publications to fall under a sin-gle pipeline promoting standardization, measurability (wherepossible) and reproducibility. The choice of documentationand data management methodology should be up to the re-searchers responsible for the reconstruction.

Several publications have discussed the taxonomy of CHrenditions to best fit with the current workflow of histori-cal experts. Foni et al. [FPMT10], for instance, proposed abroad classification system to several approaches of visualiz-ing CH objects. The taxonomy does not distinguish betweenvarious high-fidelity offline methods, which limits its use todescribe output differences for a number of physically basedalgorithms. The DECHO framework [ABV11] is a pipelinefor data acquisition, management and visualisation of CHobjects as they stand today. Our framework emphasises de-livering both physical and historical accuracy of the past.

2.2. Visualizing and assessing time and uncertainty

Visualizing time and Scene Metadata relevant to varioustime periods (including uncertainty) is an unsolved prob-

lem. Many publications addressing this topic today suggestdisplaying objects with lower confidence in less photorealis-tic graphics [RD03, ZCG05, SJW*06, SJM*10]. Several ofthese papers argue that the highlighting of Scene Metadatathrough Non-Photorealistic Rendering (NPR) approachesshows where more research needs to be done in order tominimise ambiguity for the viewer. Zuk et al. [ZCG05] sug-gest displaying temporal uncertainty in 3D models throughapproximate time windows, and by introducing uncertaintyvisual cues by rising and lowering of objects, wireframe andtransparency. A real-time 4D Virtual Reality (VR) approachwas implemented by Laycock et al. [LLDD08]. Both paperssuggest a static representation of objects that swap mod-els based on user input determining time periods, and favourblending of time periods using a slide bar representing time.

2.3. Assessing a reconstruction

Currently, the assessment of CH reconstructions remains anad hoc process, mostly without quantifiable means; makingreproducibility of any reconstruction limited. Few univer-sally accepted standards and practical issues in sharing 3Dmodels make this a recurring problem [KFH09]. Niccolucciand Hermon [NH04] proposed a fuzzy logic approach toassess possible scenarios by numerical calculating an indexof reliability, allowing for greater measurability and verifia-bility. Plecinckx [Ple08] proposed a number of methodolo-gies to develop sustainable visualisations of the past throughInterpretation Management to structure and rationalise theinterpretation process; including source assessment, sourcecorrelation and hypothesis trees, but does not critically assesshow to employ rendering methods.

2.4. Existing Predictive Rendering framework

Rendering aimed at photorealism can be broadly dividedinto two main categories: believable and predictive [Wil08].Many rendering pipelines that deliver photorealistic images,including offline approaches, fall into the former category.Predictive Rendering concerns generating images that areboth perceptually and physically correct. Its foundation isdivided into three main sections [GTS*97]: (1) Goniometric(validation of surface properties), (2) radiometric (accuratelight transport simulation) and (3) perceptual (correct out-put of final image to the human eye). For present day CHdocumentation, this framework alone provides a satisfactorypipeline to generate such results.

3. CHPR: An Overview

CHPR extends Predictive Rendering with an experimentalarchaeology approach to generate plausible renditions of thepast. A fourth comparison section is introduced to deal withthe Historical Comparison, see Figure 1. This section as-sesses accuracy of content relevant for historical experts,allowing for Scene Metadata to become a priority.

c© 2012 The AuthorsComputer Graphics Forum c© 2012 The Eurographics Association and Blackwell Publishing Ltd.

J. Happa et al. / Cultural Heritage Predictive Rendering 3

Goniometric Error Metric

Validation

Radiometric Error Metric

Radiometric Error Metric

Error Metrics

HistoricalErrorAssessment

Light Transport Simulation

Light Reflectance

Visual Display

3 10729

Historical Properties

Radiometric ComparisonGoniometric Comparison Perceptual Comparison Historical Comparison

=?

Radiometric values

Perceived image

EmissionGeometryBRDF

Plausible image

Display Observer

=?

=?

Figure 1. The framework is an extension of the existing pipeline [GTS*97]. A fourth section considers the accuracy of thehistorical data for the virtual scene. This “Scene Metadata” holds information about geometry, materials and light sourcesrelevant for its historical interpretation.

The rectangles represent measurement comparisons.Scene Metadata are important for a site’s visual appear-ance, and need to be quantified. Examples include: sun po-sition, how sunlight enters the site (e.g. coloured glass),objects that occlude areas and how they affect the lightpresent, surface properties, and finally, attributes of indoorillumination.

The square shows underlying research to process andquantify historical content in the pipeline and display thesecorrectly. All historical records available (literary texts, ex-tant objects and architecture, and images such as photographsor paintings) must be examined. Their uses must be assessedby qualified experts to aid in their interpretation.

The large arrows show that once Scene Metadata have beenadded, it can be displayed to the end observer as a plausiblerepresentation of the past. Unlike the previous three sections,however, an error metric (small arrows) cannot be objectivelyproduced from the comparison of historical records. Eachrendition must be re-assessed based on new input from his-torical experts, and then added to new iterations of assessmentuntil plausible images have been obtained. The feedback loopnature of the pipeline allows researchers to improve upon theprevious sections where necessary.

Unlike traditional Predictive Rendering, CHPR’s aim isnot only to produce results indistinguishable from the pho-torealistic ground truth. Instead, its aim is also to answerquestions relevant in CH research; particularly those thatpertain to the relationship between appearance and func-tion of a site. This is done by obtaining a number of plau-sible physically and historically based renditions to obtainnew knowledge or confirm previous assumptions about thepast.

3.1. Historical Comparison

The Historical Comparison deals with the validation of thevirtual scene as viewed by the historical expert. The lack ofquantifiable data to compare with (e.g. photographs) makesobjective assessment of most reconstructions a recurringproblem. Through cross-disciplinary collaboration, sufficientevidence (based on input available today) must be gatheredto generate renditions of the past that have scientific merit.

Once these inputs have been heuristically and experimen-tally determined, they can be added to or corrected for inthe virtual scene. To minimise error, the establishment of atimeline (starting from the site as it stands today and goingbackwards) allows for comparison of a site at its differentstages in history. Previous work mentioned in Section 2 sug-gests that timelines consisting of time windows efficientlyconvey how sites may have changed over time. CHPR adoptsa similar approach, see Figure 3.

Unlike traditional Predictive Rendering, there will not bea single correct solution, rather several Timeframes and Sce-narios need to be produced for a timeline. A Timeframe isa virtual representation of a real scene in a time window(similar to Zuk et al. [ZCG05]). However, as each of thesecan span several years, so can the visual perception withinit in each period. It can therefore consist of several Scenar-ios. Most Scene Metadata between sibling Scenarios willremain largely the same. However, a Scenario can add minorchanges; particularly how lighting alters the perception of asite. Different times of day or events may alter its appearancesignificantly, despite there not having been made any geo-metric or material changes. An example of a Scenario canbe a sermon inside a religious building, while another can bethe perception of the scene at some other time.



Figure 2. Historical Comparison consists of three mainsteps: Research Comparison, Implementation Comparisonand Historical Error Assessment. The cycle repeats until aplausible image is generated, to then determine confidenceof a historical hypothesis.

For each Timeframe, a cycle of three steps is importantto assess in order to achieve plausible output, see Figure 2 .The Research Comparison (elaborated in Sections 4 and 5)deals with the establishment of Scene Metadata for eachTimeframe and Scenario. Historical experts can both com-pare and convey expectations or known assumptions basedon input data available today to develop and update thetimeline. The present day should be the starting pointwhere possible, to then step backwards in time to theprior major alteration of the site in order to minimiseerror.

The Implementation Comparison (see Section 6) allows forthe data acquisition and processing (technological resources)available to be compared against assumptions and expecta-tions established in the previous step based on the needs ofthe reconstruction. Methods and approaches are either addedor removed from the virtual scene before rendering based ontheir relevance for the scene’s (historical) interpretation.

Once an image is synthesized, the output needs to be eval-uated in the Historical Error Assessment, see Section 7. Ifresults are deemed implausible by the historical experts, an-other round of the comparison cycle is needed, starting atthe Research Comparison again with new knowledge fromthe last step. Any physical simulation problem must be cor-rected for in the prior three sections (e.g. incorrect BRDFs,Texture Maps (TMs), etc.). Once the image is deemed plau-sible, however, it can be validated with greater confidence orrejected as a result of CHPR.

Best available tools and techniques should be used to cap-ture, process and display scenes. Resources for CH projectsmay be limited, preventing this. It is thus important to docu-ment all decisions taken before and during a reconstruction(e.g. using one of the existing standards), as well as overviewall limitations, to promote reproducibility. These notes shouldbe made throughout the whole Historical Comparison, shouldany data be lost or needed again in the future.

T , Today/ Starting Point

T , Last known significant alteration.

T , [...]

g

Figure 3. Once the primary anchor is set, T0, it is possible todetermine the last significant alteration, enabling an iterativeapproach to investigate the past. Each Timeframe requiresthe three comparison steps to ensure that a reconstructionremains plausible.

4. Research Comparison: Developing the Timeline

During Research Comparison, information regarding histor-ical resources and existing knowledge are compared to bestdetermine the starting point. If the site no longer exists, themost recent and accurate photographs, archaeological draw-ings, similar archetypes (“The original pattern or model fromwhich copies are made; a prototype” [Oxf]) or other recordsand expert opinion can be used to form the starting point.

4.1. Starting points

Generally speaking, four existing conditions can act as astarting point:

• The site no longer exists, and:– is entirely replaced by a modern site.– is mostly gone; only a few objects occupy the area.

• The site exists, and:– is extensively damaged.– is in relatively good condition.

The condition of a site determines how much data needto be gathered through experimental archaeology, and whatcan be extrapolated from existing resources, including sim-ilar sites. However, availability of (1) historical resources,(2) data capturing tools and (3) size, time, knowledge andefficiency of team working on the project are also importantfactors that make up quality of output results.

4.2. Anchoring Timeframes

Once a starting point has been established, a timeline can bedeveloped. By starting at (closest to) the present day, heredenoted by T0, it is possible to create the first iteration of thescene. This acts as a primary anchor (root node) from whichall subsequent Timeframes will be based on. If T0 existstoday, it can be validated to the real scene using traditionalPredictive Rendering, see Figure 3 .

After thorough investigation of the starting Timeframe, afundamental understanding of the environment, space and



Figure 4. Historical properties determine the appearanceof geometry, materials and light for a Timeframe. Geomet-ric changes (Occluders) may affect how much light entersa scene. Material changes affect (Filters) light reflectance.Light Attributes describe Scene Metadata about how thephysical components of the light source have been used byhuman beings.

lighting is obtained. It is then possible to go to T1; the lastknown significant alteration of the site, and from there; thesubsequent Timeframes. This is done iteratively until Tn (ori-gin or desired point in time) is reached.

All subsequent anchors are based on their ancestor Time-frames in successive order, and therefore also the primaryanchor; making T0 the most important parent. By having achild Timeframe (further back in time) inherit informationfrom its parent allows for children renditions to preserve in-formation across the entire timeline. This ensures that anyerrors that occur will at least remain consistent throughoutall Timeframes, and can be corrected for accordingly in amore straightforward, systematic manner. This also allowsfor reuse of assets and resources where appropriate, and de-creases time spent altering the scene between Timeframes.

Part of the research problem in CHPR is the attempt todevelop a logical chain of events starting from the objects’modern day state. Timelines may break into several possi-ble pasts if several hypotheses are assumed to be valid. In

such instances, a Hypothesis Tree/Graph [Ple08] is useful torepresent the timeline.

5. Research Comparison: Historical Properties

Historical properties are the Scene Metadata describing thevirtual scene (and its parameters) for a Timeframe. Determin-ing this information allows researchers to investigate how ge-ometry, materials and illumination alter the visual perceptionof a site at each Timeframe and Scenario. These properties donot have an impact on the rules of physically based simula-tion, but address how human and natural forces have affectedthe scene. The Research Comparison at each Timeframe in-volves investigation of all historical properties that make upthe virtual scene, see Figure 4.

Differentiating between physical and historical data canbe difficult if the same terminology is used, and may lead toneedlessly verbose descriptions. As Timeframes only storechanges from their parents, these alterations can be regardedas additions or subtractions to the scene (similar to versioncontrol systems). For the purpose of simplicity, Historicalproperties refer to geometric, material and lighting changesas Occluders, Filters and Light Attributes, respectively.Occluders (e.g. pillars and walls) block light entirely (orpartially) from propagating in the scene, with the addition orremoval of Filters (e.g. coloured glass, encaustics and mir-rors) providing a high-level description of scene materials.Light Attributes describe light sources (e.g. Ingredients tooil lamplight) and their uses in Scenarios and Timeframes.Figure 5 shows examples of how Historical properties canaffect a scene.

5.1. Occluders and Filters

Existing or removed geometry have features such as shapeand position. By identifying them, it is possible to investigatehow objects have blocked light, or allowed light to enter,and their purposes in relation to its parent and child Time-frame. Examples of this include neighbouring buildings,large columns, filling/removal of walls, or addition/removal

Sealed Window

Historical Properties in timeframe T- Window is sealed- No interior items- Electrical lights

Light Bulb No Glass, wooden window

Slab

Historical Properties in timeframe T- No glass, but wooden frame with slits- Different flame light ingredients (T )- Different colour slab - Candle light with T ingredients

Flame T Bricks from old hearth

Historical Properties in timeframe T- Window and glass- Indoor items (pillar, hearth, slab)- Hearth fire light (local wood)

Coloured Glass

PillarSlab

Hearth

Flame

Figure 5. Historical properties give meaningful, high-level description of a scene for its historical interpretation. Examplesinclude how the existence of glass (and type of glass/window) will affect sunlight in the scene. Interior objects can partiallyor fully occlude light sources. Surface colour and light reflectance of objects change greatly over time due to both naturalweathering and human influences. The example shows T0, Tx (a Timeframe before the origin of the site) and Tn (site origin).



Figure 6. Wall paintings inside the Red Monastery. (a) and(b) Differences in wall paintings covered in soot and afterconservation. (c) was damaged by an earthquake. (d) Paintcorrosion and external influences that have chipped off partsof the paint. Photographs courtesy of Elizabeth Bolman.

of objects. If their positioning has changed drastically, theamount of overall lighting will also differ.

Filters provide high-level contextual information on howmaterials have been added to a scene, including; changesin surface textures and BRDFs/BSSRDFs properties to de-scribe surface light reflection. Figure 6 shows photographicexamples of how wall painting surfaces can change overtime. Participating media or subsurface scattering volumes(including their shapes) are other examples of Filters, andbecome important when simulating smoke or fog. For partic-ular cases, it may be beneficial to classify the scene camera asa Filter with zoom, lens, unsharp, colourization, particularlyif there is an actual camera involved.

5.2. Rendering and illumination

Computing accurate lighting is key to deliver correct ren-ditions and minimise interpretation errors. Physically basedalgorithms today deliver the highest possible accuracy oflight transport simulation available. Unbiased approaches(e.g. point sampling methods) may have a slight advantage inthis regard as it is better to remain unbiased than to potentiallyintroduce algorithm-specific artefacts (other than variance).

Unbiased images still take a long time to synthesize com-pared to biased approaches. Fast communication betweenhistorical experts and graphics researchers is important, par-ticularly in the early days of a reconstruction project. In thefirst cycles of the Historical Comparison, it may thus be use-ful to consider physically based believable rendering in orderto use a coarse representation of the scene before refining out-put to develop plausible renditions of the past. Once an initialestablishment of the historical properties has been set, offlinerendering methods (such as unbiased algorithms) can be usedto improve the representation of the scene. At this point, ac-curacy plays a much larger role. Recently, approaches tai-lored for CH renditions have also been introduced; includingImage-Based Shooting [HBRDC11] and a system extendingSurface Depth Hallucination [MGWH11].

RTI [MGW01, FBM*06] considers image space, but isuseful in the Research and Implementation Comparisons byproviding a better understanding of CH objects. Interactive,

Figure 7. The modern day Panagia Angeloktisti rendered atvarious times of the day using locally acquired IBL.

unnatural lighting conditions such as specular enhancementsand diffuse gain can provide the information that is otherwiseinvisible to the human eye.

5.3. Light Attributes

Light Attributes refer to the high-level description of illu-mination based on historical records, expert opinion and ex-perimental data. These parameters are important in order tofully understand how the scene was observed, and how toapply them in a virtual setting. Generally speaking, lightsources can be defined as having: shape, position, directionand colour. Unlike Occluders and Filters, Light Attributescan alter the visual perception of a scene with considerablyless real-world effort between Scenarios. For instance, a sim-ple change of direction or translation of a light source canhave a significant impact on a site’s interpretation. Not onlywill the ingredients to artificial light be important, but itsuse and influence on the scene must be described for eachTimeframe and Scenarios. Light Attributes can be split intotwo main categories: Skylight and Artificial Light.

5.3.1. Skylight modelling

Sunlight has a major visual impact; from directly lit areasto sections in deep shadow. Scene illumination changes sub-stantially at different times of the day, as shown in Figure 7.Accurate simulation of the sky is computationally expen-sive as all elements need to be addressed during rendering.Simulating light transport in the sky [PSS99, JDS*01] canserve to recreate sky conditions of earlier times, when the airwas much clearer than now. Including sky elements and starconstellation can also be done if these are important for thesimulation. High Dynamic Range (HDR) imaging captureslighting taken at physical locations and can be used to vir-tually recreate natural lighting conditions at CH sites today[DTG*04, HAD*09].

5.3.2. Artificial light modelling

Artificial light can be split into two main categories: flameand electrical. Their physical simulation differ greatly, buttheir uses break down to being either for deliberate or prac-tical applications. Deliberate use of light attempts to conveya particular visual perception of a scene to the end observer.Practical use distributes as much lighting as possible in thesurroundings for visibility purposes. Careful attention must



be made to how the lights were prepared, used and main-tained to determine how to add appropriate light sources tothe scene at various Timeframes [HMD*10].

6. Implementation Comparison

The Implementation Comparison deals with evaluating ac-quired data and its subsequent processing to the expec-tations that are established in the research step, beforerendering commences. The suitability of employed tech-nologies and methodologies for the end result are deter-mined based on the needs of the historical experts. Thisincludes, for instance empirically or experimentally acquireddata, and virtually generated or processed data. This multi-disciplinary cross-checking enables for any potential misin-terpretations and miscommunications to be dealt with beforerendering.

6.1. Data acquisition

All raw data captured need to be relevant for its interpretation;including acquisition and generation of geometry, materialsand light sources, based on the historical properties. The ac-quisition of accurate experimental data is essential to developquantifiable results for a CHPR pipeline.

The starting point determined in the Research Comparisondictates the technologies needed for data collection. Subse-quent accuracy is then affected by the capabilities of thetechnologies used. This is especially the case if the raw dataare acquired experimentally, rather than generated throughhuman interpretation. Comparison of raw data with a his-torical expert may not be essential for all cases if T0 is thepresent day. Nevertheless, adding historical expert interpre-tation becomes increasingly important the further back intime Tx is.

Point cloud acquisition from laser scanning provides thehighest precision for T0 where the site still exists. Recre-ating objects through experimental archaeology techniquescan build a library of physically objects (from human in-terpretation) that can be laser scanned as well. Rome Re-born 1.0 [FAG*08] is an example of this. Procedural contentand manual 3D modelling also allow for content acquisitionthrough content generation. These data sets should be basedon archetypes, archived maps and textual records.

Furthermore, the use of spectroradiometry, for instance,will aid in collecting information about lighting. Data shouldbe based on historical records and gathered methodically,dictated by the resources available and expert opinion. Inorder to recreate a past flame, for example, it is necessary tounderstand its key components [GMMC09], such as:

(1) Ingredients: e.g. Fuel, Additives and Wick.(2) Containers: e.g. Luminaires, Torches and Hearths.(3) Surrounding environment: components of the air.

Surface colour at CH sites may be in a weathered state.While cutting a piece of a material from a sensitive site isnever an option to obtain data on site, it is often possible todetermine the as new appearance of raw materials (such asstone) with a high degree of certainty, even after millenniaof weathering by cutting out nearby similar types of materialfrom a non-sensitive site, and using this for measurement.In most cases, this should deliver satisfactory results for thequestion under discussion.

6.2. Data processing

Comparing data processing methods address how raw dataare processed in order to deliver minimal error before im-age synthesis, based on technologies available. This includesdeciding hardware tools, rendering parameters, mesh pro-cessing algorithms and surface colour estimations; i.e. theimplementation of the historical properties.

Physically based image synthesis algorithms have method-specific parameters that determine the final output. In a pathtracing renderer, termination conditions (samples per pixel,time restrictions and maximum ray depth) and image pro-cessing (filtering) images all have an impact on the finalimage. During rendering, a comprehensive description of allrendering parameters should be listed in order to provide de-tailed and reproducible results. The appropriateness of HDRand tone mapping should also be established, particularly ascareless use will lead to incorrect interpretation of the scenewith regards to colour and luminance.

Inaccuracies from simplifying Level-of-Detail (LOD) of3D models can also affect the visual perception of the sceneas light is now reflecting off a less detailed model. The weath-ered state of CH today may increase the need for TM restora-tion. Textures are difficult to reconstruct as there is no meansto validate whether the estimation is correct. Approximationsbased on existing archetypes are possible, and non-figural artmay be more straightforward in this regard. Geometric pat-terns can often be interpolated and still convey most of thesame information to the end observer. Figural art, on theother hand, have artist-specific, local and regional character-istics [Cur00, Ris01], which are important for their correctinterpretation, and may therefore be more difficult to reverse-engineer based on empirical data sets.

7. Historical Error Assessment

The Historical Error Assessment compares output imageswith the expected results from both the Research and Im-plementation Comparison steps. Historical experts can de-termine how close to the assumptions the scene rendition isand either: confirm plausibility of the synthesized image, ordeclare the image as implausible. If the image is deemedplausible, this allows historical experts to confirm or rejectprevious hypotheses/assumptions with greater confidence.On the other hand, if the image is deemed implausible, this



can be due to incorrect Physical Comparison (Goniometric,Radiometric or Perceptual) or incorrect Historical Compari-son (Research or Implementation).

7.1. Incorrect physical comparison

Incorrect physical representation of the scene results in in-correct output. Currently, materials in CH reconstruction areoften approximated as diffuse or based on analytical BRDFmodels that are heuristically applied to the scene (e.g. Phong,Ward and Lafortune). These results tend to be believable, butnot physically correct to the real scene. This can be fixed bymeasurement, however, sensitivity of CH often prevents this,resulting in slower establishment of predictive results.

If results dictate that underlying physical simulation is thereason for incorrect interpretation, the prior three sectionsmust be reconsidered. All sections can be re-evaluated be-cause of the feedback loop nature of the framework. Forinstance, in cases where scenes with high contrast betweenlight and dark areas exist; not having HDR displays to dealwith these ranges of light means that tone mapping canbe employed. However, tone mapping artefacts may affectthe interpretation of the scene. The historical experts shouldtherefore be exposed to a variety of approaches to displayingHDR content, including viewing images at a variety of singleexposures.

7.2. Incorrect Historical Comparison

Implausible images should be corrected for by establishingfeedback from both historical and graphics experts to deter-mine where the rendition may have gone wrong. It is impor-tant to keep both an open and critical dialogue to understandwhen output is implausible, and when a rendered image is infact plausible (from both a simulation and assumption stand-point), but needs to be rejected as a probable Scenario. Eachtime new information from both plausible and implausibleresults transpire, the timeline must be updated accordinglyfor the next round in the section’s cycle, or to move on to thechild Timeframe.

7.3. Assessment concerns

Another concern in this step is that historical experts com-pletely new to computer graphics may incorrectly assess pho-torealistic images. Conveying the difference between histor-ical and physical simulation accuracy is a key distinctionto make when collaborating. People unfamiliar with high-fidelity rendering (including its limitations) may, in somecases, overlook aspects of the scene. It is therefore importantduring assessment that historical experts new to photoreal-istic image synthesis to be thoroughly exposed to a varietyof rendered solutions in order to provide better judgement ofthe final image.

At the moment, the Historical Comparison will be sub-ject to expertise, hardware and resources available during

each cycle step. Each reconstruction project will set differentdemands to achieve predictive results. It is therefore im-portant to consider how external factors can influence out-put. Subjective assessment cannot be avoided, but should beminimised where possible. The evaluation can be as simpleas an informal dialogue between experts. Ideally, however,some statistical data should be generated and analyzed; eitherthrough appropriate psychophysical experiments or surveysto determine plausibility of a rendered scene. Availability ofhistorical expertise for a particular site may in some instancesprevent this.

8. Case Examples

Traditional Predictive Rendering is still not a fully solvedproblem. This is because not all real-world phenomena havebeen simulated in a physically based rendering context. CHprojects aiming for predictive results are subject to the sameand additional obstacles because the temporal domain is alsoconsidered, and not all information will be available. BothPanagia Angeloktisti [HAD*09] and The Red Monastery[Bol04, HBRDC11] were modelled using industry standard3D modelling software. The latter project was built on 2DCAD drawings developed by architectural historians on siteenabling faster modelling times.

Both models were created before the complete formulationof CHPR and contributed to its development. The case exam-ples highlight where results gave clearly visible differences,and what has been done to combat inaccurate believable im-ages. This is done instead of repeating how the models havebeen refitted to the CHPR pipeline. The iterative nature ofCHPR encourages improvement through feedback and al-lows limitations to be improved upon.

There are two primary motivations for the reconstructionsof Panagia Angeloktisti and the Red Monastery (1) to widenthe site’s understanding through high-fidelity renditions, par-ticularly as both buildings have changed their appearancesince they were first built, and (2) to document the sites asthey stand today, should anything happen to them.

8.1. Panagia angeloktisti

Panagia Angeloktisti’s origin can be traced back to the sixthcentury (T6), erected on the ruins of a three-aisled, woodenroofed early Christian basilica [Fou04], similar to the firstchurches in Cyprus [Ste10]. Its modern day appearance, how-ever, pertains largely from its 12th century iteration (T5). Ithas undergone many changes throughout history, most no-ticeably today are: (1) A vaulted chapel from the 12th centuryadded on the north side of the church (now also containingwall paintings from the 15th and 16th century). (2) A latin(Gothic) chapel added in 1302 on the south side of the church(T4). (3) Throughout the church’s history, a variety of wallpaintings and icons have been added and removed within itsspace (T3). (4) The removal of wall segments in the northern



Figure 8. Panagia Angeloktisti at the various Tx . Topleft: Old archive photo from 1955 (T1) Images courtesy ofA. Foulias [Fou04]. Top middle and right rendering of 1955.Bottom left: 1859 T2; earliest photograph known of the site,courtesy of L. Bonato et al. [BYK07]. Bottom right: Render-ing of 1859.

section, including a small tower (T2). (5) A modern Gothic-inspired bell tower that was added in the early 1900s (T1)only to be removed in 1955 as it was deemed inappropriatebased on the history of the rest of the building, by the An-tiquities Department of Cyprus. The interior has undergonea variety of changes related to additions of paintings and theplacement of interior objects [Fou04, BYK07]. Most notice-ably today is the fading and damage to the interior art andthe addition of a modern floor and chairs that largely occupythe church space.

A hypothesis graph (constructed as a tree of nodes/Timeframes) was used in the Historical Comparison to as-semble the timeline based on information from local experts,texts and local caretakers (summarized condensely above).The primary anchor is the modern day representation, fromwhich the texture maps were acquired and added for theimplementation step. As the church is still used today, it wasdecided to use the non-invasive approach of 3D modellingand site photography. Over 10 000 photographs were taken.

The visual impact of a site changes dramatically over thespan of a day. Figure 7 shows the use of image-based lighting(IBL) captured close to the real-world physical location atseveral times of the day (sunrise 05:30, 12:00 and sunset20:30).

Figure 8 shows the church at T1 (1955) and T2 (1859),respectively. Panagia Angeloktisti has not been discussed atgreat lengths in art history or archaeology literature; how-ever, it is possible to draw similarities between Panagia An-geloktisti and other churches on Cyprus such as PanagiaChrysiotissa and Panagia Afentrika [Ste10].

Figure 9. Sky model examples of Panagia Angeloktisti,12:00, 18th June 2008. Top: Outside the southern entrance.Bottom: Inside, facing the northern entrance. Each imagewas rendered for 10 h on a 24-node quad-core cluster.

8.1.1. Other considerations

Figure 9 shows that simply adding different sky models willaffect exterior and interior perception. Each of the renditionsare relit at the same time of day and year using different skymodels including IBL, the CIE Clear Sky model, Preethamand a directional light source with a background colour basedon sky values from the IBL environment map.

HDR documentation and reproduction becomes a neces-sity for scenes with significant dynamic range. As HDRdisplays are not readily available, tone mapping becomesimportant to consider. However, no Tone Mapping Oper-ator (TMO) can perfectly map all features of images forall cases. Figure 10 illustrates output from several popularTMOs [MKMS07]. The output of these images is ofvarying visual quality which may in turn affect its inter-pretation. To date, no publication investigates TMOs basedon the needs of historical experts.

During investigation of the church interior, local caretakersand historical experts argued that the perception of the mo-saic depends significantly on time of day. The considerablechange of visuals dependent on different lighting conditionsinside the Panagia Angeloktisti apse was supposed to changeits visual appearance dependent on exterior lighting.

Further investigation using RTI suggests that increas-ing the LOD for the moasic (and modelling its individual

Figure 10. Panagia Angeloktisti mid day (interior and exte-rior) with various TMOs.



Figure 11. RTI of the Panagia Angeloktisti apse. When theRTI is compared to a relit TM, clear distinctions betweeneach tesserae can be seen in the mosaic. In the TM, thescene is uniformly lit over the surface; significant perceptualcues about the scene are lost as highlighted in the VDPcomparison. RTI courtesy of Cultural Heritage Imaging.

tesseraes) is essential for its accurate representation.Figure 11 shows use of RTI relighting to investigate its vi-sual perception. The difference between simple TM relight-ing and RTI. When the RTI is compared to a relit TM, cleardistinctions between each tesserae can be seen [ZSMA07].As expected, it is observed that changes in light mainly af-fect specular materials. Areas of gold and metallic surfaceschanged the most when the lighting changes. In the TM, thescene is uniformly lit over the surface; significant perceptualcues about the scene are lost as highlighted in the VisibleDifference Predictor (VDP) [Dal93] comparison.

8.2. The Red Monastery

The Red Monastery model is still an ongoing project. Apreliminary hypothesis graph was created in the HistoricalComparison to create the timeline based on input from ex-perts and existing publications. The modern monastery actsas the primary anchor, from which TMs and 2D CAD fileswere used in the Research Comparison step. However, un-like Panagia Angeloktisti, there is significant dispute aboutthe original design of the monastery nave ceiling; one hy-pothesis being the presence of a clerestory (T6a).

Dayr Anba Bishay (known as the Red Monastery) is a Cop-tic site in Upper Egypt, near Sohag dating back to ca. 525 AD(T9). The building has an early Byzantine-inspired founda-tion and has been kept in a relatively good condition. Severalevents have shaped the site’s appearance including a catas-trophic failure of the building possibly due to an earthquake(ca. 800–900 AD, T8), reconstruction of the outer enclosurewall (ca. 900–1300 AD, T7), and the ceiling of the nave haslong since collapsed (ca. 1670, T6). The roof’s original de-sign has been a much debated topic among the experts at thesite; one hypothesis being the presence of a clerestory (T6a),and one without (T6b).

So far, on-site experts have been able to identify at leastfour phases of decorative painting including use of encaustics(hot wax painting). The wall paintings date from the sixth

Figure 12. Top: Timeframe renders of the facade. Bottom.photographs of the facade at different Timeframes. Pho-tographs courtesy of Nicholas Warner and the Red MonasteryProject/American Research Center in Egypt.

to the seventh or eighth century (Late Antiquity)[Bol04].The soot inside the sanctuary was not caused by domesticactivities, but in celebration of the mass, with oil lamps,tallow candle and incense probably contributing the most tothe obscuration of the wall paintings.

Since then, a small village occupied within the nave be-tween 1700 and 1890 (T5), before it was partially clearedin 1909 (T4). Medieval renovations, a 20th century facade(added 1910–1914, T3) and a dome on the eastern sectionhave contributed to the preservation of the triconch sanctuary(and its wall paintings) [Bol04]. The remainder of the villagewas cleared between the 1930 and 1960 (T2). A full interven-tive conservation project was initiated at the Red Monasteryin 2002 (T1). Today, conservation is nearing completion (T0).Figure 12 shows example output results of the exterior naveat the various Timeframes (T2, T3 and T4), while Figure 13shows a proposed 2D CAD drawings of the past monastery,with the clerestory (T6a).

8.2.1. Other considerations

Similarly to Panagia Angeloktisti, sky models have an im-pact on the visual perception of a scene. The monastery has

Figure 13. 2D CAD model of the Red Monastery. Im-age courtesy of Nicholas Warner and the Red MonasteryProject/American Research Center in Egypt.



Figure 14. Path traced sky model examples of RedMonastery (T0), showing different amount of indirect light.

considerably less access from which light may enter. This isparticularly noticeable in some sky models over others dueto the surrounding wall around the nave blocking light fromentering on ground level. Lack of access to areas of lightmakes a considerable visual impact for some scenes, makingit appear darker overall, see Figure 14.

As mentioned, it may be advantageous in the first iter-ations of the CHPR feedback loop to consider physicallybased believable rendering methods to fast obtain clear gen-eral overview of illumination in order to refine and developplausible renditions. For such instances, using techniquesthat employ empirically captured illumination may be use-ful. Image-Based Shooting (IBS) [HBRDC11] used HDRphotography to accelerate image synthesis of the interiormonastery for previewing purposes, see Figure 15. As esti-mating BRDFs is not straightforward for such sites, and phys-ical measurement is not an option, the previewing methodallows for simple previewing, closer to the ground truth thanwhat diffuse path tracing is able to deliver. Furthermore, itwas also shown to be useful to aid CH documentation and3D model validation to reality.

9. Discussion

Barring the invention of a time machine, or some unforeseentechnology, it will not be possible to determine the original

Figure 15. Comparing photograph, IBS and path tracing.

appearance of structures with 100% certainty. CHPR maythus appear as a contradictory term because of the equivo-cal nature of historical reconstruction. Speculative contentshould be permissible if the goal of the investigation can bemet in the presence of such content. For several sociologicaland archaeological problems, this may be the case. This isunlike traditional Predictive Rendering, where total accuracyis the sole purpose of the whole effort.

9.1. Limitations of case studies

Gonioreflectometers are still experimental devices. Theirwidespread use is not likely in the near future. This is amajor problem as correct light reflectance models rely on ac-curate measurement. No portable gonioreflectometers existin a format that is usable at CH sites. For our case studies,matte materials were approximated as diffuse BRDFs, whichis the case for a large number of CH models [HMD*10]. Allother glossy and specular materials such as the gold paintedicons and the tiled floor in Panagia Angeloktisti were ap-proximated using analytic BRDF models such as Ward andLafortune [NDM05].

Despite its recognized status as an important historical site,few scholarly publications have been published on PanagiaAngeloktisti, limiting the comprehensiveness of our timeline,and lack of branching timeline. The correct modelling ofgold mosaics to the specifications of Byzantine environmentshas not been implemented, despite their importance havingbeen demonstrated through RTI. The Red Monastery, on theother hand, is currently undergoing significant interventiveconservation, with information about the site being updatedfrequently, which subsequently has been the case for the 3Dmodel as well. In the future, we wish to investigate reversingthe weathering of the encaustics, and further investigate thenave roof’s design.

9.2. Limitations of CHPR

Predictive Rendering is largely a theoretical research frame-work, developed from technology that is still experimental,14 years after the original publication. Most graphics groupstoday do not have the facilities to address all sections of thepipeline. Data acquisition and portability are not straightfor-ward processes either. This may improve as hardware ad-vances in the future.

There are instances in which the full pipeline may not benecessary to investigate the past. Devlin et al. [DCB02], forexample, described the relighting of cave art from 15 000years ago, suggesting that the careful crafting of a horse tothe cave wall may have been used to create crude anima-tions of a horse’s gait using dynamic flames. In this instance,CHPR would be able to produce a T0 for the present day anda T1 with reconstructed points added to the T0 point cloudbased on expert opinion. However, little evidence suggestsmajor damage at several Timeframes, making subsequent



Txs largely unnecessary. No additional information can beextrapolated than what T0 can already deliver through phys-ical simulation. In this case, only the historical properties ofthe flame and its use are relevant from CHPR.

10. Future Work and Conclusion

In this paper, we have proposed an extension to the PredictiveRendering pipeline to study CH in the present and the past.There are several existing efforts towards standardization inthis research domain [3dC09, Gro96, DCM95, AG07, Eur08,IFL09, DH09, GB11, ISO10]. To our knowledge, however,no existing ontology has addressed how to best employ ren-dering techniques to display CH in both the present and past.As CHPR embodies both Scene Metadata and rendering tech-niques, we believe that a number of these standards, however,can be employed simultaneously, within the various steps ofthe Historical Comparison forming predictive results.

Like Predictive Rendering, CHPR is not a fully solvedproblem. Not all necessary information will be available forinterpretation. The pipeline should therefore solely be re-garded as a research framework in the attempt to minimiserendering error before, during and after image synthesis. Oursolution suggests an iterative and progressive approach basedon resources available, but also allows for corrections and up-dates to be added in a straightforward manner.

Acknowledgements

The authors thank the Byzantine Art Gallery, IoannisEliades, Andreas Foulias and the caretakers of Panagia An-geloktisti. They also thanks the ARCE, Elizabeth Bolman,Nicholas Warner, Agnieszka Szymanska, Laurel D. Hackley,Gerry D. Scott III and Michael Jones, Elmedin Selmanovic,Carlin Yuen and Red Monastery project helpers. Tone Map-pers are based on source code by Mantiuk et al. [MKMS07].CIE Clear Sky and Preetham sky are from Banty’s toolkit[Ban06]. Images in Figure 6 courtesy of Elizabeth Bolman.Figure 13 courtesy of Nicholas Warner. RTI in Figure 11courtesy of Cultural Heritage Imaging.

References

[3dC09] 3DCoFORM: 3dcoform website. http://www.3d-coform.eu/, 2009. Accessed December 2011.

[ABV11] ALIAGA D. G., BERTINO E., VALTOLINA S.: Decho: Aframework for the digital exploration of cultural heritageobjects. Journal on Computing and Cultural Heritage 3,3 (2011), 1–26.

[AG07] ARNOLD D., GESER G.: Research Agenda for theApplications of ICT to Cultural Heritage. EPOCHPublications, 2007. http://public-repository.epoch-net.org/publications/RES AGENDA/research agenda.pdf.

[Ban06] BANTERLE F.: Banty’s toolkit. http://www.banterle.com/francesco/projects.php, 2006. Accessed February2012.

[Bol04] BOLMAN E.: The red monastery conservation project,2004 campaign: New contributions to the corpus of lateantique art. Interactions: Artistic Interchange Between theEastern and Western Worlds in the Medieval Period, 2004.

[BR02] BERNARDINI F., RUSHMEIER H.: The 3d model acqui-sition pipeline. Computer Graphics Forum 21, 2 (2002),149–172.

[BYK07] BONATO L., YIAKOUMIS H., KABA K.: The island ofCyprus. Nicosia, Cyprus, 2007.

[CL10] CONYERS L., LECKEBUSCH J.: Geophysical archae-ology research agendas for the future: Some ground-penetrating radar examples. Archaeological Prospection17, 2 (2010), 117–123.

[Cur00] CURCIC S.: Middle Byzantine Architecture onCyprus: Provincial or Regional? The Bank of Cypurs Cul-tural Foundation, Nicosia, Cyprus, 2000.

[Dal93] DALY S.: The visible differences predictor: An algo-rithm for the assessment of image fidelity. Digital Imagesand Human Vision 4 (1993), 2–15.

[DCB02] DEVLIN K., CHALMERS A., BROWN D.: Predictivelighting and perception in archaeological representations.In UNESCO “World Heritage in the Digital Age” Pro-ceedings of 30th Anniversary Digital Congress, UNESCOWorld Heritage Centre (2002).

[DCM95] DCMI: Dublin core metadata initiative. http://dublincore.org/, 1995. Accessed December 2011.

[DH09] DENARD H., HERMON S.: London charter. http://www.londoncharter.org/, 2009. Accessed December 2011.

[DTG*04] DEBEVEC P., TCHOU C., GARDNER A., HAWKINS T.,POULLIS C.: Estimating surface reflectance properties of acomplex scene under captured natural illumination. USCICT Technical Report ICT-TR-06.2004, 2004.

[Eur08] EUROPEANA: Europeana portal. http://www.europeana.eu/, 2008. Accessed December 2011.

[FAG*08] FRISCHER B., ABERNATHY D., GUIDI G., MYERS J.,THIBODEAU C., SALVEMINI A., MULLER P., HOFSTEE P., MINOR

B.: Rome Reborn. In Proceedings of ACM SIGGRAPHComputer Graphics: New Tech Demos (2008).

[FBM*06] FREETH T., BITSAKIS Y., MOUSSAS X., SEIRADAKIS

J., TSELIKAS A., MANGOU H., ZAFEIROPOULOU M., HADLAND

R., BATE D., RAMSEY A., ALLEN M., CRAWLEY A., HOCKLEY

P., MALZBENDER T., GELB D., AMBRISCO W., EDMUNDS



M.: Decoding the ancient greek astronomical calculatorknown as the antikythera mechanism. Nature 444 (2006),587–591.

[Fou04] FOULIAS A.: The Church of our Lady Angeloktisti atKiti. The mosaic decoration of Our Lady Angeloktisti atKiti. Submitted as an Master’s thesis to the Department ofHistory and Archaeology, University of Cyprus, Nicosia,Cyprus, 2004.

[FPMT10] FONI A. E., PAPAGIANNAKIS G., MAGNENAT-THALMANN N.: A taxonomy of visualisation strategies forcultural heritage applications. Journal on Computing andCultural Heritage 3, 1 (2010), 1–21.

[GB11] GRANDE A., BENDICHO V.: The principles of thesevilla charter. In Proceedings of the XXIVth InternationalCommittee for Documentation of Cultural Heritage Sym-posium (2011).

[GMMC09] GONCALVES A., MAGALHAES L., MOURA J.,CHALMERS A.: High dynamic range—A gateway for pre-dictive ancient lighting. Journal on Computing and Cul-tural Heritage 2, 1 (2009), 1–20.

[Gro96] CIDOC DOCUMENTATION STANDARDS WORKING

GROUP: CIDOC Conceptual Reference Model. http://www.cidoc-crm.org/, 1996. Accessed December 2011.

[GTS*97] GREENBERG D. P., TORRANCE K., SHIRLEY P., ARVO

J., FERWERDA J. A., PATTANAIK S., LAFORTUNE E., WALTER

B., FOO S.-C., TRUMBORE B.: A framework for realisticimage synthesis. In Proceedings of the ACM SIGGRAPHComputer Graphics (1997).

[HAD*09] HAPPA J., ARTUSI A., DUBLA P., BASHFORD-ROGERS

T., DEBATTISTA K., HULUSIC V., CHALMERS A.: The virtualreconstruction and daylight illumination of the PanagiaAngeloktisti. In Proceedings of the VAST InternationalSymposium on Virtual Reality, Archaeology and CulturalHeritage (2009).

[HBRDC11] HAPPA J., BASHFORD-ROGERS T., DEBATTISTA K.,CHALMERS A.: Rendering interior cultural heritage scenesusing image-based shooting. In Proceedings of Euro-graphics, Area Paper (Heritage) (2011).

[HMD*10] HAPPA J., MUDGE M., DEBATTISTA K., ARTUSI

A., GONCALVES A., CHALMERS A.: Illuminating the past:State of the art. Virtual Reality 14, 3 (2010), 155–182.

[IFL09] IFLA INTERNATIONAL FEDERATION OF LIBRARY

ASSOCIATIONS AND INSTITUTIONS: Functional require-ments for bibliographic records. http://www.ifla.org/publications/functional-requirements-for-bibliographic-records/, 2009. Accessed December 2011.

[ISO10] ISO INTERNATIONAL ORGANIZATION FOR

STANDARDIZATION: ISO 690:2010, Guidelines for biblio-graphic references and citations to information resources.http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?csnumber=43320, 2010. AccessedDecember 2011.

[JDS*01] JENSEN H. W., DURAND F., STARK M., PREMOZE S.,DORSEY J., SHIRLEY P.: A physically-based nightsky model.In Proceedings of the ACM SIGGRAPH Computer Graph-ics (2001).

[KFH09] KOLLER D., FRISCHER B., HUMPHREYS G.: Researchchallenges for digital archives of 3d cultural heritage mod-els. Journal on Computing and Cultural Heritage 2, 3(2009), 1–17.

[LLDD08] LAYCOCK S. D., LAYCOCK R. G., DRINKWATER D.,DAY A. M.: Real-time visualisation of multiple time de-pendent reconstruction hypotheses for a cultural heritagesite. In Proceedings of the VAST International Symposiumon Virtual Reality, Archaeology and Cultural Heritage(2008).

[MGW01] MALZBENDER T., GELB D., WOLTERS H.: Polyno-mial texture maps. In Proceedings of the ACM SIGGRAPHComputer Graphics (2001).

[MGWH11] MELENDEZ F., GLENCROSS M., WARD G.,HUBBOLD R.: Relightable buildings from images. In Pro-ceedings of the Eurographics, Area Paper (Heritage)(2011).

[MKMS07] MANTIUK R., KRAWCZYK G., MANTIUK R., SEIDEL

H.-P.: High dynamic range imaging pipeline: Perception-motivated representation of visual content. In Proceedingsof the Human Vision and Electronic Imaging XII (SPIE,2007).

[NDM05] NGAN A., DURAND F., MATUSIK W.: Experimentalanalysis of BRDF models. In Proceedings of the Euro-graphics Symposium on Rendering (2005).

[NH04] NICCOLUCCI F., HERMON S.: A fuzzy logic ap-proach to reliability in archaeological virtual recon-struction. In Proceedings of the CAA Computer Ap-plications and Quantitative Methods in Archaeology(2004).

[Oxf] OXFORD ENGLISH DICTIONARY: “Archetype” definitionas found in the oed. http://www.oed.com/. AccessedDecember 2011.

[Ple08] PLECINCKX D.: Interpretation management: Howto make sustainable visualisations of the past. InProceedings of the Conference on Open Digital CulturalHeritage Systems (2008).



[PSS99] PREETHAM A., SHIRLEY P., SMITS B.: A practical ana-lytic model for daylight. In Proceedings of the ACM SIG-GRAPH Computer Graphics (1999).

[RD03] ROUSSOU M., DRETTAKIS G.: Photorealism and non-photorealism in virtual heritage representation. In Pro-ceedings of the VAST International Symposium on VirtualReality, Archaeology and Cultural Heritage (2003).

[Ris01] RISEBERO B.: The Story of Western Architecture (3rdedition). Herbert Press, Herbert Press, London, 2001.

[SJM*10] SIFNIOTIS M., JACKSON B., MANIA K., VLASSIS N.,WATTEN P. L., WHITE M.: 3d visualisation of archaeologicaluncertainty. In Proceedings of the ACM Symposium onApplied Perception (2010).

[SJW*06] SIFNIOTIS M., JACKSON B., WHITE M., MANIA K.,WATTEN P.: Visualising uncertainty in archaeological re-constructions: A possibilistic approach. In Proceedings

of the ACM SIGGRAPH Computer Graphics: Sketches(2006).

[Ste10] STEWART C.: The first vaulted churches in Cyprus.Journal of the Society of Architectural Historians 69, 2(2010), 162–189.

[Wil08] WILKIE A.: Predictive Rendering. Habilitation the-sis, Vienna University of Technology, 2008.

[ZCG05] ZUK T., CARPENDALE S., GLANZMAN W. D.: Visualiz-ing temporal uncertainty in 3D virtual reconstructions. InProceedings of the VAST International Symposium on Vir-tual Reality, Archaeology and Cultural Heritage (2005).

[ZSMA07] ZANYI E., SCHROER C., MUDGE M., CHALMERS

A.: Lighting and byzantine glass tesserae. In Proceed-ings of the EVA Electronic Visualisation and the Arts(2007).


cultural heritage predictive rendering

Documents