Potential and Challenges of using Computer Graphics for the Simulation of Optical Measurement Systems

Max-Gerd Retzlaff1,2 (retzlaff@kit.edu), Johannes Hanika1, Jürgen Beyerer2,3, Carsten Dachsbacher1 1Karlsruhe Institute of Technology (KIT), Institute for Visualization and Data Analysis (IVD),

Computer Graphics Group, Karlsruhe, Germany 2Fraunhofer Institute of Optronics, System Technologies and Image Exploitation (IOSB),

Karlsruhe, Germany 3Karlsruhe Institute of Technology (KIT), Institute for Anthropomatics and Robotics,

Vision and Fusion Laboratory (IES), Karlsruhe, Germany

Abstract Physically-based image synthesis methods, a research direction in computer graphics (CG), are ca-pable of simulating optical measuring systems in their entirety and thus constitute an interesting ap-proach for the development, simulation, optimization, and validation of such systems. In addition, other CG methods, so called procedural modeling techniques, can be used to quickly generate large sets of virtual samples and scenes thereof that comprise the same variety as physical testing objects and real scenes (e.g., if digitized sample data is not available or difficult to acquire). Appropriate image synthe-sis (rendering) techniques result in a realistic image formation for the virtual scenes, considering light sources, material, complex lens systems, and sensor properties, and can be used to evaluate and improve complex measuring systems and automated optical inspection (AOI) systems independent of a physical realization. In this paper, we describe an image generation pipeline for the evaluation and optimization of measuring and AOI systems, we provide an overview over suitable image synthesis methods and their characteristics, and we discuss the challenges for the design and specification of a given measuring situation in order to allow for a reliable simulation and validation.

Keywords: computer graphics, procedural modeling, image synthesis, optical measurement

Introduction Current physically-based image synthesis techniques constitute a major leap compared to previously used, mostly phenomenological approaches. The simulation of light transport is at the core of physically-based image synthe-sis methods and crucial to generate images that are on par with images made by physical image acquisition systems. Light transport simulation nowadays is almost exclusively computed using Monte Carlo (MC) or Markov Chain Monte Carlo (MCMC) methods, which can account for complex light-matter interac-tions and naturally handle spectral emission, absorption, and scattering behavior (measured or derived from models) described by geomet-ric optics. (MC)MC methods can also comprise the simulation of complex lens systems to ac-curately compute the resulting irradiance onto a virtual sensor. Essentially, all (MC)MC rendering methods compute an estimate of the light transport by sampling, that is, stochastically generating, paths on which light propagates from light sources to sensors; their main difference being the path sampling strategy. Until sufficient

convergence, the variance of this estimation is apparent as noise in the images. Because of this, even simplistic realizations of these meth-ods are versatile and, in principle, capable to achieve the desired, and required, results. However, their application is only practical when an (MC)MC method is used which is well-suited for a given scenario; otherwise the computation time can easily become prohibi-tively long, even in seemingly simple cases. For example, one would choose different methods for computing light transport for com-plex high-frequency light transport phenomena (recognizable by multiple glossy or specular reflection) than for highly scattering media. Particular attention has to be paid to the simu-lation of complex lens systems which can in-crease the computation time by orders of mag-nitude when implemented naively. We discuss the use of a state-of-the-art approach to effi-cient rendering with realistic lenses in the con-text of measuring and AOI systems. As indicated, many different rendering algo-rithms and sampling strategies exist in the realm of (MC)MC methods, and they all exhibit different performance and noise characteris-tics, which are strongly linked to the type of

light-matter interactions and the geometry con-figurations occurring in a scene. As such, it is not straightforward for the non-expert to select the appropriate method. For this purpose, we identify light interactions and phenomena con-stituting different challenges for the image synthesis, and discuss state-of-the-art algo-rithms, as, for example, path tracing, bi-directional path tracing (BDPT), and others.

Rendering, rasterization, and ray tracing Hughes et al. [8] define the term rendering very concisely as referring to the process of integra-tion of the light that arrives at each pixel of the image sensor inside a virtual camera in order to compute an image. There are two major strategies for determining the color of an image pixel: rasterization and ray tracing. Ray tracing, also sometimes referred to as ray casting, determines the visibility of surfaces by tracing rays of light from the virtual view point, that is, the viewer’s eye or the image sensor, to the objects in the scene. The view point repre-sents the center of origin and the image a win-dow on an arbitrary view plane. For each pixel of the image a view ray is sent originating in the view point through the pixel into the scene in order to find an intersection with a surface. By recursive application of this ray casting one can compute complex light interactions and global illumination, that is, indirect illumination including, among others, reflections and shad-ows. Rasterization, on the other hand, projects geometric primitives one by one onto the im-age window. A depth buffer, also called z-buffer, is utilized to determine the closest and thus visible primitive for each pixel. Usually, the perspective projection is carried out in three steps. First, the projection trans-formation, expressed in homogeneous coordi-nates, is applied. Afterwards, the projective coordinates are dehomogenized by the nor-malization transformation, mapping the view frustum to the unit cube. The resulting device coordinates can then be mapped to image coordinates by discarding the depth coordi-nate, as it is done for orthographic projection. All primitives can be processed in parallel us-ing a single instruction multiple data (SIMD) approach and minor synchronization via the depth buffer. This allows for a very fast pipe-lined hardware implementation in the form of modern graphics processing units (GPUs). Simplified, one could say that ray tracing starts with the pixels and then determines ray inter-sections with the scene geometry, while raster-ization starts with the geometry, projecting it onto the image plane. The availability of mod-ern GPUs makes rasterization feasible for interactive real-time application.

Synthetic images in the context of optical inspection In [19] we describe the idea of using computer graphics methods to allow systematic and thorough evaluation of automated optical in-spection (AOI) systems. Instead of using real objects and acquisition systems, computer graphics methods are used to create large virtual sets of samples of test objects and to simulate image acquisition setups. We use procedural modeling techniques to generate virtual objects with varying appearance and properties, mimicking real objects and sample sets. Physical simulation of rigid bodies is de-ployed to simulate the placement of virtual objects, and using physically-based rendering techniques we create synthetic images. These are used as input to an AOI system instead of physically acquired images. This enables the development, optimization, and evaluation of the image processing and classification steps of an AOI system independently of a physical realization. We demonstrated this approach for shards of glass, as sorting glass is one challenging prac-tical application for AOI. In this paper, we focus on the aspects of image synthesis.

Real-time rendering of shard distributions Our implementation includes real-time render-ing of shard distributions by hardware rasteri-zation on a GPU using OpenGL 4.2 [21]. As glass is an optically semi-transparent mate-rial, a method of transparency rendering is necessary. OpenGL itself only provides alpha blending that can be used to render transpar-ent materials. But rendering transparency us-ing alpha blending implies rendering the ob-jects in sorted order. As sorting for every rendered frame is impractical and even not always possible (e.g., for mutually overlapping objects) order-independent transparency ren-dering (OIT) techniques have been invented. Our previous implementation as presented in [19] used depth peeling (introduced by Everitt [2]), a robust hardware-accelerated OIT method, but meanwhile we replaced this with an implementation of a more powerful OIT technique making use of per-pixel concurrent linked lists (introduced by Yang et al. [24]). OIT rendering using depth peeling can store only a quite limited number of material interactions per image pixel requiring multiple rendering passes in case of many interactions. Per-pixel linked list OIT, on the other hand, allows stor-ing a high number of interactions in a single rendering pass, and the realistic rendering of the breaking edges of glass shards may re-quire a high number of interactions as the edg-es can be quite rough. As our previous publications [19] and [20] fo-cus on the generation of realistic virtual objects and scenes based on measured data as well

as the mimages fwe had tion andthesis, eimage foMost iming. Theonly threblue nospectral

Replicatand speRecentlymethod replicateacquisitiothe simuresponseLight emas full sdata of rinteractiosorption reflectionresultingplied witfunction,across thues of ean RGBlated senIn casetions [22describeas standcolorimeconverteexampleplayed oserve asdesignincolor imsimulateSimulatiocount fo

Fig. 1:

methodical prfor the evaluused simplif image acqu

e.g., ideal senormation. portantly, we

e synthesis ee values drm stimuli, effects such

ting real imaectral rendery, we haveto support

e more physon systems,

ulation of reaes of real im

mission and tspectra. Thareal, measuron is compuspectra (of c

n spectra ag spectral ph a color ma, as depicthe spectrumach sensor s triplet in thensor that can

e of the C2] XYZ coorde the color pedardized by

etric observeed to an RGe, the sRGBon a calibras a referenceng a AOI setumpression oned viewing coon of chrom

or the differe

Emission halogen lamnation Tech

rocedure of uuation of realfied models ouisition for thnsors and sim

e have usedwas therefo

describing reand could

h as dispersio

age acquisitring enhanced spectral renical aspects , that is, weal light sourcage sensorstransport we

at is, we usred light souuted using course, for o

are equally power distrib

atching or seted in Fig.

m leads to thesensitivity fue RGB spacn be displaye

CIE color mdinates are erception of ay the CIE 1r. The XYZ i

GB working B color spacated screene image for aup as it correne would peonditions, cf. matic adaptence in white

spectrum omp (Model 3hnologies, In

using synthel AOI systemof the illuminhe image symplified optic

d RGB rendeore reduced ed, green, anot reproduon.

tion system

our real-timndering and

of real imae now suppoces and sens.

e now descrie the spectrces. The ligmeasured ather use caspossible). Tution is munsor sensitiv2. Integratie intensity vnction, usua

ce of the simed. matching fun

computed tha human bei1931 standaimages can space as, f

ce [9] and dn. It thus can expert usesponds to terceive in tFig. 4. ation can ae reference

of a tungst3900 by Illumnc.).

etic ms, na-yn-cal

er-to nd

uce

ms

me to ge ort sor

be tral ght ab-ses The ulti-vity on

val-ally mu-

nc-hat ng

ard be for

dis-can ser the the

ac-of

ten mi-

Fi

thpu

LiWbuspsoexdecoimcamdeupchillOimthconaFobithinartrabismesde

Fi

ig. 2: RawUC4

he simulateduter screen t

ight sourcesWhile light so

ulbs or halopectra, see ources as, xhibit a numepicted in Fiombination w

mages in wannot be pe

more. This caering, and hep that doeshoosing a suumination, s

Of course, smportant in thhe light specombined withation. or our real-inning appro

he full spectrunto bins of are approximaa can either ins, which imooth spectssary in theepicting narr

ig. 3: Emisceilin

w sensor res4/UC8 line sc

light emisshat is used f

s with brighurces such aogen lamps Fig. 1, the sfor example

mber of narrg. 3. Such a

with certain shich actual

erceived or an be simulatelps to desig

s not exhibituitable combee Fig. 5 andspectral renhe context ofctrum to a nh lighting by

-time implemach to spectum of the visa certain widated by a ste

be approxims a sufficienra, or it can e case of ilow bright pe

ssion spectrng light.

sponse of thcan camera b

sion and of for viewing.

ht spectral lias incandes

have quitespectra of oe, fluorescerow bright pa lighting cosensors canspectral dif

easily detected with spegn an acquist such probination of sed compare to

ndering is ef color filtersarrow specta non-unifo

mentation wtral renderingsible light is qdth, thus theep function. Tmated by eqnt approximbe adaptive

llumination weaks.

rum of a flu

he ELiiXA by e2v.

the com-

nes cent light

e smooth other light nt lamps

peaks, as ndition in result in fferences cted any-ectral ren-sition set-blems by ensor and o Fig. 4. especially that limit ral range rm illumi-

we use a g. That is, quantized e spectra The spec-ual width

mation for e, as nec-with light

uorescent

Fig. 4:

PhysicaWhile it complexexamplein fact beimplemeaccept tone poinmore feaSupport zation aply, it is mray traccase. Buexceedetion of rmatic an

GeneratRetzlaff for glassscribed balgorithmcertain sly relocathe glassvolume while wedisplacefaces usshaders These min negliging surfaa downsment, thintersectintersectSmall irr

Images rethe color mCIE 1931 server. Toptom: fluores

ally-based reis possible

x light interae, refraction een done in

entation for ligthat rasterizant a ray tracasible, and eof refractionpproach by r

much more stcer. Dispersut the afore

ed when gloeal lens sys

nd chromatic

tion of synthet al. [19] des shards baby Martinet em to generatsmoothing, aated and this s shards. In ais smooth t

e achieve plament mappising hardwathat were int

modifications gible computaace detail in side. In casehe polygonations or holetion surface regularities c

ndered in rmatching fustandard co

p: CIE Illumiscent illumin

endering to add sup

action phenoand dispersform of a pr

ght refractionation has itsing approachven more eff

n can be addray marchingtraightforwar

sion constituementioned pbal illuminat

stems includaberration is

hetic glass sescribe a proased on an et al. [17]. Wte plausible

as waste glasleads to rou

addition to thto allow fasausible breang of the inre-acceleratetroduced in O make a shaation time poreal-time. Bu

e of consideral mesh ms at the tranand original

can be fixed

real-time usinctions of tolorimetric oinant D65, bonation of Fig.

pport for moomena as, fion – and hroof of concen –, one has s limits and h begins to ficient. ed to a rasteg but, obviourd to add it toutes a simipoint is suretion or simuing monochrs asked for.

shards ocedural mod

algorithm dWe modified t

shards withss is repeateunded edges his, our carvist intersectioaking edges tersection sued tessellatiOpenGL 4.0.ard generatiossible by adut there is alrable displac

might get sesition betweshard surfaceasily but b

ing the ob-ot-3.

ore for

has ept to at be

eri-us-o a lar ely la-ro-

del de-the h a ed- of ng

on, by ur-on . on

dd-lso ce-elf-en ce. ig-

Fi

gepocoscanbaofsuThgethinshMcotiotiosuthouFrmOduVanmretagrregrraatmge

CloInm

ig. 5: Imagthe Edepiimag

er defects wolygon meshonsiderably. cribe surfacend thus prased renderif conservatiouch irregularherefore, weenerating sh

he requiremenstead introdhard meshes

Most existing omputer grapons or tesseons generateurfaces that hus not suitaut remeshingracture imp

modeling softwOne notable

uced by Kororonoi partitnd emphasiz

meshing of thegions are nance but instraph definedegions can braph edges.andom noisete irregular,

making it a senerating of

omparison o rendering tn all their dmethods shar

ges rendereELiiXA sensicted in Fig. ges illuminate

would requireh, slowing d

Such erroes that are noesent problng methods

on of energy ities. e vouched fohards of broent of real-tducing a pos. fracturing m

phics are baellations, bue only undetaare also notble for surfag. An examplementation ware by the Bexception i

rndörfer [16].ioning but uzes on achiee boundary

not based ontead on the by the voxebe controlled The metho edge weighi.e., plausibleuitable candvirtual glass

of the suitatechniques

diversity, there the conc

ed in real-timsor sensitivity

2. Top andted as in Fig.

e a remeshindown the geoneous mesot physically lems for p

s. As a resultmight get v

for a new moken glass, time generaool of prec

methods in thased on Vorout many impailed, flat intt well-tessellace perturbaple for this is

of the BleBlender Fouis the meth. It is also buses a 3D veving a regusurface. Then the Euclidlength of p

els. The shad by weighod already hts in order e, surface st

didate for oushards.

bility of Monfor AOI

e (MC)MC rcept of stoch

me using y function d bottom 4.

ng of the eneration shes de-possible, hysically-t, the law iolated at

method of dropping

ation and computed

he field of onoi parti-plementa-tersection lated and tion with-s the Cell ender 3D ndation. od intro-

based on voxel grid ular quad e Voronoi dean dis-aths in a pe of the ts of the supports to gener-tructures, ur task of

nte Car-

rendering hastically

Fig. 6:

creating lights. Indering aexist in tPath tracamera, action, o(NEE, Kplement from smtions to tLight trasources,ing, aga(here: colematic lens modonly emiBi-directpaths arlight soubetweenvia multpracticalmethod well in stions arethe casenate objand lightMany-liglight sublights. Thdiffuse sthe downfaces anular surfaPhoton mnode lopinges otion to eworks wFor glospath trarobust aimages estimatioror), cf. H

A synthetigenerated erated by oframework.

paths, connn this sectionalgorithms anthe realm of (cing: The pdirections a

optionally, wKajiya [13]). P

and suitablmall light sou

the light souracing: Paths , direction sain optionally onnection to in context dels. It workit to the surfational path tre starting frources, with

n nodes. It reiple importal, which unon-trivial tocenes wheree not blockee for AOI setects which at source.

ght methods bpaths and this approachscenes and nside, it has

nd is even imfaces, cf. Dacmapping: Crecations (loc

on surfaces),estimate en

well with diffusy surfaces,cing. Photo

and has thewith low noon also causHachisuka an

ic image ovirtual glass

our Monte C.

necting the n, we discussnd sampling (MC)MC meaths are sta

are sampled with next evPath tracing e if there arces and if rces are posare starting

ampling as wwith next evcamera). Buof specular

ks only well iaces visible ttracing (BDPom both enddeterministi

equires variance samplin

unfortunately o implemente the determed. This shottings as we are visible to

(variants of Breat nodes a

h works very is easy to i problems w

mpossible to chsbacher eteate light su

cations wher and use a dergy per su

use surfaces it (gracefullyn mapping

e advantageise levels, bses a bias (nd Jensen [5

f proceduras shards ge

Carlo renderi

sensor to ts different restrategies ththods.

arting from tat every inteent estimatiis easy to ire no caustidirect connesible. from the lig

with path travent estimatiut NEE is pro

surfaces aif light sourcto the cameraPT): The sus, camera ac connectioance reducting (MIS) to

makes tht. BDPT wor

ministic conneuld usually want to illum

o both came

BDPT): Creaas virtual powell for mosmplement. O

with glossy suuse with spet al. [3]. bpaths, recore energy imdensity estimurface area. s and causticy) degrades can be ma

e of producibut the dens(systematic e5].

ally en-ing

the en-hat

the er-on m-ics ec-

ght ac-on

ob-nd

ces a. ub-nd

ons on be his rks ec-be mi-era

ate oint stly On ur-ec-

ord m-

ma- It cs. to de ng

sity er-

Fi

MmModalplpainspsiMspgrMvetraerinthofthnecufoprexco

BFiw

Fi

ig. 7: A cldispangl

More extensiomethods existMetropolis lig

ds use tlgorithm (intrle the spaceaths. Numer

n primary sppace (Veachons such as

Manifold Exppace transpradient doma

MLT methodsery powerfuansport pher, they have

ndependent hus rely on thf said light p

his results in ents are renurrences ofound over timroperty that txception isontrol.

ringing it alig. 6 shows

with the meth

ig. 8: ClosLeft:brighFig.D652048

lose-up viewlayed in Figle.

ons and varit, as, e.g., Geht transport he Metroproduced by He of all posous variantspace (Kelemh [23]), as wes Energy Disloration (Jakort (Kaplany

ain methods s have in coul in explnomena (e.gto be initia

MC samplerhose to detecphenomena. images whedered with lthe light p

me. Also, MLthey are diffiKelemen [1

l together a scene of t

hod of Korn

se-up view o: fluorescenht spectral l3, right: CIE. Both image8 samples pe

w of the scg. 6 from a

iants to bi-deorgiev et al(MLT): The

polis-HastingHastings [7])ssible light

s exist, e.g., men [11]) orell as modestribution Rakob [10]), Hayan et al. [1(Kettunen et

ommon that loring difficg., caustics)

alized repeatrs (e.g., BDct actual occ In image reere individualittle noise bphenomena LT methods sicult to imple1]), and d

the shards gdörfer [16] u

of a synthetnt illuminatilines, as deE Standard Ies are rendeer pixel.

cene also different

irectional . [4]. se meth-s (MH) ) to sam-transport sampling

r in path rn exten-aytracing, alf-Vector 14]), and t al. [15]). they are

cult light . Howev-tedly with DPT) and currences endering,

al compo-ut all oc-are only

share the ement (an ifficult to

generated using our

tic shard. ion with

epicted in Illuminant ered with

Fig. 9:

existing supportincally-basinherentreflectionpath-tracAOI setuAs the spectral tral rendpling, it cclose-up(from a refractioIt is inteilluminataffect theform samto noisy narrow more hosamplestrum. ThImportansamplingwhile stleveragesamplingrenderin

RealisticIn interaperspectto the pring systproacheslens modeven a c

Synthetic ition (bottomlens modelof field (DO

Monte Carng global illused simulatioly accounts ns and shacing without up exhibits a ray tracing binning but

dering with Mcan simulatep view of theslightly differn and chrom

eresting to notion and intere rendering mpling of theimages in c

peaks compomogenous

s fall equally his is illustratence samplingg rate in regill being an

e this probleg count (ang time).

c lenses active imagetive projectiorojection of atems also s as the pidel. While it

complex lens

image and m) renderedl exhibiting a

OF).

rlo ray traciumination, thon of light tra

for phenomadows. We

NEE for ouvery large ligframework instead supp

Monte Carlo dispersion.

e same scenrent angle),

matic dispersiote that the racting surfatime or imag

e emission scase of specpared to im

spectral illon every pa

ed in Fig. 8. g that resulgions with h

unbiased eem without nd thereby i

e synthesis,on is used thaa pinhole caoften use nhole cameis possible

s into the virtu

enlarged sed using a tha narrow dep

ng framewohat is, a phyansport, whimena such

have chosur task as tght source.does not uports full spespectral sa

Fig. 7 showsne as in Figexhibiting ligon. spectra of tces might al

ge quality. Uspectrum leactra with brig

mages usingumination,

art of the spe

ts in a highigher radianestimation cincreasing tincreasing t

often simpat corresponmera. Rendesimplified ara or the thto just inclu

ual scene an

ec-hin pth

ork ysi-ich as en

the

use ec-m-s a . 6

ght

the lso ni-

ads ght

a as

ec-

her nce can the the

ple nds er-ap-hin de

nd

Fi

tracitranothWseporecoelenexFicosisyFiththlelethstfeTovico

Fi

ig. 10: Samsimping chro

ace rays thrent renderinacing, 95% ot leave a re

he actual sceWe use our mented in Hanort efficient eal world lensompared to rl. Such real ns of lens example, a Dig. 11. Still, ompute the mulate flawsystems. ig. 9 and Fig

he same scehin lens modens (achromenses have ghe image bytrong chromerred to as coo conclude tew of rendeolor matching

ig. 11: Dounieu

me scene as ple Lensbaba less narr

omatic aberra

rough it, thisng, as, for of all ray sameal world lenne [6]. method and nika and Dacrendering usses with onlyrendering wiworld lenseslements andouble-Gauss

this approlens image s and shortc

g. 10 show ane but usingdel and a si

matic doublegot different

y the Lensbamatic aberraolor fringing. this section, erings with g and sensor

ble-Gauss leux [1]

in Fig. 9 buby-like lensrow DOF buation (color fr

s leads to vestraight forw

mples, or mons system a

implementachsbacher [6sing a wide y negligible oith the thin les can consisd an apertures lens as deoach can aand thus becomings of

a group of img different leimple Lens

et), respectivt depths of faby-like lensation, comm

we present different lenr sensitivity f

ens by Pier

ut using a s, exhibit-ut strong fringing).

ery ineffi-ward ray ore, might and enter

ation pre-6] to sup-range of

overhead ens mod-st of doz-e, as, for epicted in accurately e used to real lens

mages of nses, the

sbaby-like vely. Both field, and s exhibits

monly re-

an over-nses and functions,

rre Angé-

Fig. 12:

Fig. 13:

Fig. 14:

Fig. 15:

respectivfrom a fla top viearrangemtem. Ac

Synthetic imlens mode1931 standCIE D65 he

Same as Angénieux also depicgrid is overin distortion

Same as ithe ELiiXA

Same as F

vely. Fig. 12 lat angle, whew of the sament of the rtually, in the

mage using l as perceivdard colorimemispherical

in Fig. 12 Double-Ga

cted in Fig. rlaid to maken easily reco

in Fig. 13 busensor, dep

Fig. 13 but wi

to Fig. 15 shile Fig. 16 toame scene rreal image ae real syste

the simple thved by the Cetric observel illumination

but using tauss lens [11. The bla

e the differengnizable.

ut detected icted in Fig.

ith motion blu

how the sceo Fig. 19 shoresembling tacquisition syem the shar

hin CIE ver, .

the [1], ack nce

by 2.

ur.

ne ow the ys-rds

alinAcaboscanofas

CInsypoimanThVIt nobeofthreul

R[1

[2

[3

[4

[5

[6

[7

[8

[9

lways appean Fig. 18 and s we generaal simulationodies, we ccenes separnd thus alsof the shardss to be seen

onclusion an summary, ynthesis pipeonents as w

mplementationd validationhe EMVA ision [12] deis dissecte

oise or, more described f dark noise

hermally geneadout noiselate the sens

References] Pierre Angé

nent Opticaissued Feb

] Everitt Castransparencration, May

3] Carsten DaHašan, AdaNovák. ScaLight Metho33(1):88–1

4] Iliyan GeorgDavidovič, simulation wACM TransGRAPH Asdoi: 10.114

5] Toshiya HaStochastic Trans. GrapAsia 2009, doi: 10.114

6] Johannes HEfficient Molenses. Comrographics)doi: 10.111

7] W. Keith Haods using MBiometrika doi: 10.109

8] John F. HuFoley, StevPrinciples aWesley, 20

9] IEC (Internasion). MultiColour mea2-1: Colourspace – sR

ar in a strongFig. 19.

ate the virtua of gravity a

can easily grated by smo render ima

sliding alonin Fig. 15.

and future wwe describe

eline includiell as their e

on with rega of optical m1288 Stan

scribes a moed as a comre generally,as the squagoverned b

nerated elect. This model

sor noise of r

énieux. Large al Objective. U. 15, 1955. s. Interactive ocy. Technical

y 2001. achsbacher, Jaam Arbree, Bralable Realisticods. Compute04, 2014; doi: giev, Jaroslavand Philipp Slwith vertex cos. Graph. (TOGsia 2012, 31(6)5/2366145.23

achisuka and Hprogressive pph. (TOG) – P28(5):141:1–85/1618452.16

Hanika, and Conte Carlo renmputer Graph), 33(2):323–31/cgf.12301 astings. MonteMarkov chains57(1): 97–1093/biomet/57.1ghes, Andries

ven K. Feiner. and Practice, 315; ISBN: 978ational Electromedia systemasurement andr management

RGB. IEC 6196

g transmitted

al scenes usiand collisiongenerate conmall time diffages with mong a sloping

work ed the entirng all relevefficient and ard to the s

measuring sysndard for odel for sensmbination o, shot noise are-root of thby the dark ctrons, and ol can be usereal sensors.

Aperture Six U.S. Patent 2,

order-indepenreport, NVIDIA

aroslav Křivánruce Walter, ac Rendering w

er Graphics Fo 10.1111/cgf.1

v Křivánek, Tolusallek. Light

onnection and G) – Proc. of S):192:1–10, N

366211 Henrik Wann J

photon mappinProc. of SIGGR8, Dec. 2009; 618487

Carsten Dachsndering with rehics Forum (Pr332, April 2014

e Carlo sampls and their app9, April 1970; .97

s van Dam, JaComputer Gr

3rd edition. Ad8-0-321-39952otechnical Com

ms and equipmd management – Default RG66-2-1:1999; I

d light as

ng physi-s of rigid nsecutive ferences, otion blur

g surface,

re image ant com-accurate

simulation stems. Machine

sor noise. of photon

that can he signal, current of of sensor ed to sim-

Compo-701,982,

ndent A Corpo-

nek, Miloš nd Jan

with Many-orum, 12256 máš

t transport merging.

SIG-ov. 2012;

Jensen. ng. ACM RAPH

bacher. ealistic roc. of Eu-4;

ing meth-plications.

ames D. aphics: dison-2-6 mmis-

ment – nt – Part GB colour SBN: 2-

8318TC 1

[10] Wenzexploniquelar traof SIGdoi: 1

[11] CsabAntalbust Trans(Proc2002

[12] Berndchine2010

[13] JameSIGG150,

[14] AntonCarstReprLight (TOG33(4)doi: 1

[15] MarkJaakkZwickTrans2012doi: 1

[16] JohaFestkUnive

[17] Auréland Scrackling In349,

[18] JohaCarstMarkScaleterialSIGGdoi: 1

[19] Max-Beyeing imtomanischGruyt

[20] Max-sten and pinspetung ing, 2

[21] MarkGrap(CoreApril

[22] Thomorimetions 1931

[23] Eric VLight Stanf

8-4989-6; ICS 00 zel Jakob andoration: a Marke for renderingansport. ACM GGRAPH 20110.1145/21855ba Kelemen, Ll, and Ferenc Mutation Stratsport Algorithmc. of Eurograp; doi: 10.1111d Jähne. EMVe Vision. Optik0; doi: 10.1002es T. Kajiya. TGRAPH CompAug. 1986; don S. Kaplanyaten Dachsbacesentation of Transport Sim

G) – Proc. of S):102:1–13, Ju10.1145/26010kus Kettunen, ko Lehtinen, Fker. Gradient-s. Graph. (TO, 34(4):123:1–

10.1145/27669nn Korndörferkörpern in Bruersität Karlsrulien Martinet, Samir Hakkouks and fracturenternational (S2004; doi: 10.nnes Meng, Mten Dachsbac

kus Gross, ande Modeling ans. ACM Trans

GRAPH 2015, 10.1145/27669-Gerd Retzlafferer, and Carstmages using pated optical inshes Messen, 8ter; doi: 10.15

-Gerd RetzlaffDachsbacher.procedural moection (AOI) sy2014, pages 4

2014; doi: 10.5k Segal and Kuphics System: e Profile) – Ap2012.

mas Smith andetric Standardof the Optical–32; doi: 10.1Veach. RobusTransport Sim

ford University

codes: 33.160

d Steve Marsckov Chain Mog scenes with

Trans. Graph2, 31(4):58:1–

520.2185554 László SzirmayCsonka. A Simtegy for the Mm. Computer Ghics), 21(3):53/1467-8659.t0

VA 1288 Standk & Photonik, 52/opph.20119

The rendering puter Graphicsoi: 10.1145/15an, Johannes Hcher. The Natuthe Path Spac

mulation. ACMSIGGRAPH 20uly 2014; 097.2601108 Marco Manzi, Frédo Durand,domain path tG) – Proc. of

–13, Aug. 201997 r. Interaktive Z

uchstücke. Diphe (TH), 2011Eric Galin, Bre

uche. Procedues. Proc. of thShort Paper), .1109/SMI.200

Marios Papas, cher, Steve Mad Wojciech Jand Rendering os. Graph. (TOG

34(4):49:1–13949 , Josua Stabeten Dachsbac

parameterizedspection (AOI)82(5): 251– 61515/teme-2014, Josua Stabe. Synthetic imaodeling for autystems. Forum47–59, KIT Sc5445/KSP/100urt Akeley. ThA Specificatio

pril 27, 2012).

d John Guild. Ts and Their Ul Society 33(3088/1475-487

st Monte Carlomulation. Ph.Dy, 1998; ISBN

0.60, 37.080 –

hner. Manifoldnte Carlo techdifficult specu

h. (TOG) – Pro–13, July 2012

y-Kalos, Györgmple and Ro-

Metropolis LighGraphics Foru31–540, Sep. 01-1-00703 dard for Ma-5(1):53–54, 0082 equation. ACM

s, 20(4):143–5886.15902 Hanika, and ural-Constraintce for Efficient

M Trans. Grap014),

Miika Aittala, , and Matthiastracing. ACM SIGGRAPH 5;

Zerlegung vonploma thesis. 1. ett Desbenoit,ral modeling oe Shape Modepages 346–04.1314524 Ralf Habel,

arschner, rosz. Multi-of Granular MaG) – Proc. of 3, Aug. 2015;

enow, Jürgen cher. Synthesiz models for au). tm – Tech-, April 2015. D

4-0041 enow, and Carage acquisitioomated optica

m Bildverarbeicientific Publis00043608 e OpenGL

on (Ver. 4.2 Khronos Grou

The C.I.E. Cose. Transac-):73–134, 78/33/3/301 o Methods for D. dissertation: 0-591-90780

–

d h-u-oc. 2;

gy

ht um

M

t t h.

s

, of el-

a-

z-u-

De

r-on al -

sh-

up,

l-

. 0-1

[2

Fi

Fi

Fi

Fi

4] Jason C. Yand NicolaLinked List the 21st Euing (EGSR’doi: 10.111

ig. 16: AngCIE CIE

ig. 17: SamELiiX

ig. 18: Samcenttranssphe

ig. 19: Samsens

ang, Justin Has Thibieroz. R

construction ourographics Sy’10), pages 121/j.1467-8659

énieux DouD65 hemis

standard col

me as Fig. 16XA sensor, d

me as Fig. 16t ceiling lighsmitted ligherical illumina

me as Fig. 18sor, depicted

Hensley, HolgReal-Time Conon the GPU. Pymposium on297–1304, 20

9.2010.01725

uble-Gauss spherical illulorimetric ob

6 but detectedepicted in F

6 but with theht of Fig. 3 aht as well aation.

8 but with thd in Fig. 2.

ger Grün, ncurrent Proc. of n Render-010;

lens [1], umination, bserver.

ed by the Fig. 2.

e fluores-as strong as hemi-

he ELiiXA