Image Processing for the Optimisation of Dynamic Range in Photomicrography

Transcription

1 I MAGE P ROCESSING Imge Processing for the Optimistion of Dynmic Rnge in Photomicrogrphy Jörg Piper, Medun-Klinik, Bd Bertrich, Germny BIOGRAPHY Jörg Piper otined his medicl doctorte from the University of Bonn, Germny. One field of his scientific ctivities hs een directed to light microscopy pplictions, especilly reflection contrst, luminnce contrst nd digitl imge processing. He hs creted severl mthemticl models for 3D quntittive nlyses of opque cellulr specimens. Since 1998 he hs collorted with the University of Orde, Romni, s n ssocite nd honorry professor. In 2008 he ws ppointed s reserch professor nd n honorry den y the New Europen University (UNE), Brussels, Belgium. In Germny, he currently works s senior consultnt for internl medicine, ngiology nd dietology. ABSTRACT Dynmic rnge increse (DRI, synonym: exposure lending) nd high dynmic rnge rendering (HDR) re oth suitle techniques for improving the imge qulity in photomicrogrphs, especilly with specimens or illumintion modes tht re ssocited with lrge vrition in oject rightness. To chieve this, the specimen hs to e photogrphed t different exposures, nd the respective imge stck hs to e superimposed nd rendered y prticulr softwre. The resulting imge reconstruction is free from ny visile over- or underexposed zones. In this rticle, vrious softwre solutions re descried with specil regrd to microscopicl pplictions nd compred with ech other. KEYWORDS light microscopy, photomicrogrphy, dynmic rnge, illumintion, imge processing, rendering, Photomtix, Picturenut, Esy-HDR, DRI-Tool, Imge Stcker ACKNOWLEDGEMENTS The uthor thnks Mr Eerhrd Rp for some preprtions of ditom shells. AUTHOR DETAILS Prof. Dr med. Jörg Piper, Clinic Medun, Clr-Vieig-Strsse 4, D Bd Bertrich, Germny Tel: +49 (0) Emil: jpu.mp@t-online.de We: Microscopy nd Anlysis 23(1):S5-S9 (EU),2009 INTRODUCTION In oth everydy photogrphy nd photomicrogrphy, lrge vritions in oject rightness my cuse technicl prolems in cpturing fithful imge. In such cses, only limited rnge of luminnce will e properly exposed wheres righter nd drker res will e overexposed or underexposed, respectively. In photogrphy, this prolem rises in ll nlogue nd digitl cmers when the photogrphs re tken in the usul wy (i.e. single shot). They re relted to the phenomenon tht the humn eye is cple of perceiving much higher rnge of luminnce thn ny photogrphic film or sensor, so humns nturlly tend to think they cn lwys get well exposed picture in dim or right light. Thus, for instnce, difficulties will regulrly occur in everydy photogrphy when the interior of church hs to e photogrphed in nturl light. When the rightly lit church windows re well exposed, the other res will e underexposed so tht ny detil inside the church will not e visile; conversely, when the drk zones within the uilding re well exposed, the windows will e extremely overexposed. More severe difficulties my occur when prtilly illuminted ojects hve to e photogrphed in drkness. In light microscopy, photomicrogrphy is ffected with similr limittions when specimens or illumintion modes re ssocited with high rnges in contrst nd luminnce, especilly in oservtions crried out using drkfield illumintion, polrized light or fluorescence techniques. Figures 1 to 1d show some typicl high-contrst imges of utterfly wing illuminted in polrized light nd tken t different exposures of 2 s to 1/250 s (equivlent to 9 exposure vlues, or EV). This clerly shows tht it is not possile to crete single stisfctory imge of this specimen showing ll relevnt detil within the very right chitin vessels nd the much drker scles. Dynmic Rnge The rnge of luminnce of n oject or imge is clled the dynmic rnge nd is defined s the rtio of the mximum to the minimum vlues of luminnce within the specimen or imge. Luminnce is mesured in lux or cndels per squre metre (1 lux = 1 cd m -2 ). Tle 1 gives some typicl vlues for luminnce with regrd to severl nturl scenes. It shows tht scene showing the interior of room within sunlit view outside the window, for instnce, will hve dynmic rnge of pproximtely 100,000:1 [5]. When the dynmic rnge of n imge is 10,000:1 or lower, it is defined s low dynmic rnge imge (LDR) [10-12]. Dynmic rnges higher thn 10,000:1 re clled high dynmic rnges (HDR). On typicl disply screens (e.g. lptop, computer, TV, etc.), only low dynmic rnges re present (out 1:100 to 1:500). In ll types of photogrphic prints, the dynmic rnge is much lower (circ 1:32 to 1:64) [1]). In digitl cmers, the mximum dynmic rnge detectle y the sensor is out 1:1000. When n nlogue (film) cmer is used, dynmic rnges up to 1:10,000 cn e recorded y the film. In prctice, the rnge of drkness nd rightness within specimen cn lso e quntified y the EV. For instnce, the rnge of exposure is 3 EV when the drkest region within n imge is well exposed y 1/30 s nd the rightest y 1/250 s. It is very importnt for prcticl photogrphic strtegies tht the mximum rnge in rightness should e lower thn 2 EV when optimum qulity is desired for prints. In high dynmic rnge constelltions, the resulting exposure time is usully clculted s the men vlue of the mximum nd minimum times corresponding with the rightest nd drkest zones. However, loss of visul informtion will result from this. Fundmentl improvements to the visul informtion cn e chieved when n imge sequence is mde using different exposure times. In this wy, n imge stck cn e creted consisting of vrious imges showing different regions of the specimen in optiml exposure. Imge Processing for Dynmic Rnge In such situtions, the qulity nd visul informtion in photogrphed imges cn e optimised y using two types of imge processing clled high dynmic rnge rendering (HDR) nd dynmic rnge increse (DRI). For exmple, Figure 1e shows corresponding HDR reconstruction of Figures 1-d mde using the imge processing softwre Photomtix Pro; it shows much more detil thn ny single imge of the originl stck, nd over- or underexposed zones re no longer visile. Severl softwre solutions re ville for imge reconstruction in high dynmic rnge situtions. Five of these products, either freewre or shrewre nd suitle for microscopicl pplictions, were rigorously tested in this study. The sl principles of their opertion, their specific fetures nd the results of comprtive softwre tests crried out using different specimens nd illumintion conditions re reported in this rticle. Principles of Dynmic Rnge Rendering When set of imges is creted consisting of severl imges tken t different exposures, S5

2 the set cn e rendered into single, optiml imge in two different wys: y high dynmic rnge rendering (HDR); or y dynmic rnge increse (DRI, synonym: exposure lending). In HDR techniques, the three colour chnnels of red, green nd lue, nd the lph chnnel corresponding to the trnsprency, re trnsformed from 9 to 32 its. Moreover, ll tonl vlues re coded s floting point dt. In this wy, the numer of tonl vlues cn e enlrged from 256 (s usul in 8-it imges) up to round 4.3 illion grdtions per chnnel. From this high dynmic rnge imge is reconstructed with, in most cses, dynmic rnge etween 10,000:1 nd 100,000:1. Such HDR imges cn neither e printed out or oserved on usul screen in stisfying mnner ecuse their dynmic rnges re too high. Therefore, HDR imges hve to e trnsformed into new LDR imges; this seprte step in imge processing is clled tone mpping. When tone mpping is dequtely crried out, the resulting finl imge shows the mximum shrpness nd detil of fine structure over the full rnge of luminnce, nd is free from ny visile over- or underexposed zones. When DRI (exposure lending) is crried out, the tonl vlues remin constnt in ll single imges; HDR imges re not creted nd tone mpping is not necessry. The softwre extrcts ll well exposed prts (or zones) directly from the imge stck nd composes new imge. Thus, pre-existing lrge differences in rightness nd contrst cn e reduced. c e d MATERIALS AND METHODS The following softwre solutions were tested using severl types of specimens: DRI-Tool nd Imge Stcker [3,4], Esy HDR Pro nd Bsic [7], Picturenut [2,6] nd Photomtix Pro [5,9]. All specimens were illuminted in high contrst modes using drkfield or polrized light microscopy. From ech specimen, imge sequences were tken t vrying exposures s descried ove. The differences in exposure of two successive single imges rnged etween 1 nd 2 EV; these spcings re mostly recommended y other uthors s well s y the softwre producers. The numer of imges per sequence ws dependent on the dynmic rnge of the scene since the luminnce vlues of ll detils hd to e reproduced, regrdless of whether they were drk, midtones or highlights. All imges were cptured with n Olympus Cmedi C megpixel digitl cmer mounted on Leic lortory microscope equipped with pln chromtic nd pochromtic ojectives, Vrio photo-oculr , nd Zernike universl condenser for rightfield nd drkfield, polriztion nd phse contrst. For the ditoms shown in Figures 3-5, verticl stcks were tken in order to enhnce their depth of field efore the HDR renderings were performed. The different stcking softwre used for this were Comine Z 5, Picoly nd Helicon Focus. Thus, deep-focus imges were reconstructed eing different in their expo- Figure 1: (-d) Four polrized light microgrphs of the wing of utterfly (lmd compenstor, rnge of exposure: 9 EV). A representtive qurtet of single imges t different exposures. Horizontl field width (HFW) = 2 mm. (e) High dynmic rnge (HDR) reconstruction using Detils Enhncer of the utterfly wing shown ove using 10 single imges t different exposures. Figure 2: Drkfield light microgrphs of the hed nd thorx of gnt. Rnge of exposure: 4 EV. () Single imge. () HDR reconstruction using Tone Compressor from five single imges. HFW = 1.5 mm. S6

3 sure, nd completely free from ny relevnt indistinctness. The techniques necessry for this hve lredy een descried in previous pper [8]. In second step, the creted deepfocus imge sequences were trnsformed into HDR imges using Photomtix Pro. SOFTWARE SOLUTIONS All the tested softwre solutions were cple of trnsforming the finl reconstructed LDR imges into the stndrd imge formts (JPEG, TIFF nd BMP). The key chrcteristics of the severl products re compiled in Tle 2. DRI-Tool nd Imge Stcker These freewre products [3,4] for DRI rendering hve minor differences in their design nd sizes ut they gve identicl results. When the single imges were selected from the respective imge stck, the softwre creted reconstructed DRI imge utomticlly. Thus, the mode of imge processing cnnot e influenced y the user. The resulting reconstructions cn only e modified when different single imges were preselected for rendering. Esy HDR Bsic This freewre [7] is cple of creting HDR imges which cn e mnully trnsformed into new LDR imges. The chrcter of the reconstructions cn e modified y the user with the help of preview tool. Thus, severl relevnt prmeters cn e djusted when the tone mpping is crried out (e.g. grdtion, histogrm, white nd lck point, gmm, sturtion); dditionl imge filtering is lso implemented (Gussin lur, shrpen, medin nd ilterl filter, colour djust, nd white lnce). Esy HDR Pro The shrewre version of esy HDR [7] is dditionlly fitted with msk mode so tht the strength of tonl equliztion, the intensity of contrst compression, rightness, smoothness nd edge rendering cn e djusted. As result the finl imges my e more lnced thn with Esy HDR sic. Picturenut Picturenut [2,6] is n HDR-sed freewre eing nerly comprle with Esy HDR Pro. For HDR-rendering, this softwre offers two vrint modes, the dptive logrithmicl mode nd the photoreceptor physiology mode. In the first mode, the following prmeters cn e mnully modified while the tone mpping is crried out: exposure correction, equliztion of light nd shdow, men rightness, contrst nd histogrm-sed grdtion. In similr mnner, the other rendering mode llows modifictions of exposure correction, dynmic compression, colour sturtion, contrst nd histogrm-sed grdtion. According to our experience, oth vrints led to comprle results. Photomtix Pro Photomtix is the only softwre designed for HDR s well s for DRI techniques. Thus, n imge stck cn e rendered in different I MAGE P ROCESSING Figure 3: Drkfield light microgrphs of ditom shells. Rnge of exposure: 3 EV. () Single imge. () HDR reconstruction using Detils Enhncer from four single imges. HFW = 0.5 mm. Figure 4: Polrized light microgrphs of the ditom shells shown in Figure 3 (crossed polrizers, qurter lmd compenstor, rnge of exposure: 4 EV). () Single imge. () HDR reconstruction using Detils Enhncer from five single imges. HFW = 0.5 mm. S7

4 Figure 5: Ditom shell in drkfield illumintion with simultneous trnsmitted monochromtic lue light nd red epi-illumintion (rnge of exposure: 1.5 EV). () Single imge. (): HDR reconstruction from two single imges. HFW = 0.15 mm. modes so tht the est result cn e selected. In the HDR technique, tone mppings cn e processed in two different wys, sed on the mcros Detils Enhncer or Tone Compressor. When Detils Enhncer is used, mximum numer of prmeters cn e mnully influenced y the user: strength of contrst enhncements, colour sturtion, light smoothing, luminosity, white nd lck point, gmm, colour temperture, sturtion of highlights nd shdows, micro-contrst, micro-smoothing, contrst enhncements in highlights nd shdows, nd shdow clipping. Tone Compressor works more simply, so tht lower numer of prmeters cn e modified in the tone mpping procedure: rightness, tonl rnge, contrst dpttion, white nd lck point, colour temperture nd colour sturtion. For reconstructions sed on DRI severl mcros re implemented. Three mcros exist for utomtic running: Averge (glol verging), Highlights nd Shdows for two imges, nd Highlights nd Shdows for more thn two imges. Both the ltter mcros led to enhnced contrst in the drk nd right zones. For mnully controlled DRI reconstructions, two seprte mcros re ville: Highlights nd Shdows Adjust nd Highlights nd Shdows Intensive. In the djust mode, the lending point (weight given to the overexposed versus underexposed imges) nd rdius (djustment of shrpness) cn e regulted. In the intensive mode, the intensity of tonl equlizing cn e pre-selected (light or enhnced); moreover, the precision of imge rendering cn e selected, so tht potentil hlo rtefcts my e reduced more precisely. RESULTS OF SOFTWARE TESTS DRI-Tool nd Imge stcker cn e strted from the desktop or ny other dt crrier, e.g. USB stick. They worked very fst, so lrge numer of imges cn e rendered in firly short time. Additionlly, oth progrms led to the lowest equliztions in right nd drk zones when compred with the other softwre solutions. Moreover, in contrst to the other softwre, single imges selected for rendering cnnot e redjusted or centred y these tools. Esy-HDR Bsic nd Pro gve superior results ecuse severl relevnt prmeters cn e individully djusted. When the rnge of rightness nd contrst ws very high, the msk mode of the Pro version my furthermore contriute to fundmentl improvements in shrpness nd tonl lnce. In most cses, Picturenut seemed to e comprle with Esy HDR Pro. However, in contrst to the other progrms, this softwre does not work without the originl EXIF dt from the single imges. Thus, Picturenut cnnot e used for reconstructions when the respective single imges were preprocessed y softwre tht removes their EXIF dt. Photomtix Pro promised the most impressive results t lest for most microscopicl pplictions. According to our experience, HDR rendering should e preferred s more detils re usully recognizle thn in DRIsed reconstructions. The Tone Compressor might e n dvntge in rre cses if the Detils Enhncer should produce too much detil which does not relly exist in the respective specimen. Otherwise, in most routine situtions, reconstructions crried out with the help of Detils Enhncer were the est when compred with corresponding imges creted y other mens. Therefore, ll Figures presented here were mde with Photomtix Pro. Detils Enhncer enles the user to crete mximum, wide-rnging equliztions in high contrst imges; moreover, high differences in rightness cn e trnsformed into different colours eing nerly equl with regrd to their luminnce. Especilly y this mens, fine detils cn e enhnced lso in very drk or right regions nd the finl imges cn e oserved or printed t the highest qulity. Figure 2 shows gnt in drkfield; there is very lrge difference in rightness etween the right ntenns nd the drk cput nd thorx. In this cse, the Photomtix Tone Compressor led to etter results thn the Detils Enhncer. Within the specimen, the reconstructed imge shows complete equliztion in rightness; in prticulr, the originl drk structures occur with improved clrity nd luminnce. Figures 3 to 6 demonstrte the visily improved effects chievle y the Photomtix Detils Enhncer in typicl high contrst imges tht differed in size, opticl density nd nisotropy (they were illuminted with drkfield or polrized light). It shows tht Detils Enhncer is lso le to equlize differences in the rightness of the specimens nd the drk or lck ckground. Moreover, the luminnce of colours is improved. Figure 5 shows ditom illuminted simultneously with polrized light, trnsmitted lue light nd epi-illumintion red light. In this typicl imge, the rnge of exposures ws rther low (1.5 EV). Figure 6 shows polrized light imges of scoric cid crystls. In oth Figures 5 nd 6 the HDR reconstruction show much more detil, especilly with regrd to fine structures or textures within the specimen. Moreover, the shrpness of fine detil is visily enhnced, nd regionl looming is eliminted very effectively. CONCLUSIONS When differences in luminnce re greter thn round 1 to 2 EV, photomicrogrphic S8

5 I MAGE P ROCESSING imges cn mostly e optimized y DRI or HDR techniques. Suitle softwre cn e successfully used for high contrst illumintion modes. As DRI tools crete only moderte improvements in most cses, HDR-sed imge processing should e preferred t lest in microscopicl pplictions. According to ll tests crried out, Photomtix Pro in prticulr seems to e the superior softwre in ll relevnt fields of light microscopy. The Detils Enhncer will mostly promise the est results; in rre cses, the lterntively usle Tone Compressor might led to etter results if the Detils Enhncer should produce rtificil pseudo-detils or if nturlly existing detils eing rely visile in relity should e enhnced in n unnturl mnner. Not only re the differences in specimen rightness equlized nd fine detils enhnced ut lso the clrity of colours cn e visily improved, especilly with the help of Detils Enhncer. Moreover, the glol contrst cn e reduced when n originl drk ckground is modified so tht it ppers with higher rightness nd luminnce. When n up-to-dte computer is used, the rendering process is finished within few minutes so tht HDR nd DRI techniques could esily e integrted into microscope workflows. Overll, suitle softwre solutions for HDR imging might e useful for ll microscopists interested in optimized digitl imge presenttions nd optimized print documenttion. REFERENCES 1. Cspri, C. CCInfo: Advice for ville light photogrphy in night scenes (in Germn). nchtufnhmen.htm 2. Dougherty, J. As good s the eye cn see ( review out picturenut). s_good_s_the_eye_cn_see 3. Gross, S., Hrtl, B. Enhncements of visile dynmic rnges using the DRI-Tool softwre, 2008 (in Germn) 4. Hrtl, B. Imge Stcker, 2008 (in Germn). /Dten/ImgeStcker.exe 5. HDR Soft: Photomtix, Mehl, M. Picturenut, HDRI-genertor, tone mpping, 2008 (in Germn) Okonek, B. esyhdr: Alwys properly exposed imges y exposure lending nd tone mpping, Piper, J. Use of softwre to enhnce depth of field nd improve focus in photomicrogrphy. Microscopy nd Anlysis 113:15-19, Ssso, F. How to crete HDR-photos - HDR/Photomtix tutoril. Tle 2: Some chrcteristics of the tested softwre. Figure 6: Crystls of scoric cid in polrized light with lmd compenstor (rnge of exposure: 5 EV). () Single imge. () HDR reconstruction from six single imges. HFW = 1.0 mm. hdrphotomtix-tutoril 10. Spilker, H. HDR-photogrphy more relity thn in relity. Spiegel Online networld, 2007 (in Germn) Wikipedi: High Dynmic Rnge Rendering, en.wikipedi.org/wiki/high_dynmic_rnge_rendering 12. Wikipedi: High Dynmic Rnge Imging, John Wiley & Sons, Ltd Tle 1: Luminnce of some typicl nturl light sources or scenes Light Source or Scene Luminnce (cndel m -2 ) Strlight (10-5 ) Moonlight Living room 1.0 Outdoor shde 100 Outdoor sunlit 100,000 Sunlight 10,000,000 (10 7 ) Softwre Progrm Size (MB) Instlltion Required Cost Methods Shde Depth Imge Formts Min. Operting System Mnul Interctive Processing DRI Tool 1 no freewre DRI s single imges TIFF Win 95 no Imge Stcker 0.97 no freewre DRI s single imges TIFF Win 95 no Esy HDR Bsic 1.36 yes freewre HDRI 24 it TIFF, JPEG, BMP Win 95 yes Esy HDR Pro 1.75 yes ~ 50 US$ HDRI 24 it TIFF, JPEG, BMP Win 95 yes Picturenut 4.37 no freewre HDRI 8, 16 nd 32 it TIFF Win 2000 yes Photomtix Pro 13.9 yes ~ 150 US$ DRI 8 nd 16 it TIFF, JPEG, BMP Win XP, Mc 05 yes S9