designing camera origination films for scan-only

SMPTE MMoottiioonn IImmaaggiinngg Journal, April 2006 • wwwwww..ssmmppttee..oorrgg 112211

As an integral component in the television or digi-tal intermediate process, a scan-only film mustbe designed to deliver superior digital data

when compared to a traditional optically printable film.Such quality can be delivered in the form of improvednoise structure (lower grain), higher spatial resolution,extended dynamic range, or enhanced color reproduc-tion. Further, the additional benefits of traditional filmimage capture such as format independence, equip-ment compatibility, and archivability, among others,must not be compromised. When paired with intelligentdigital image processing, the superior capture capabili-ties of the scan-only film can be used to deliverimproved creative and technical content.

An obvious application for such a system would be torender the classic color and tone reproduction charac-teristics of any existing traditional film, but with qualityenhancements above and beyond what is available inthe more classic workflows such as direct optical or con-tact printing. Beyond traditional camera origination filmemulation, scan-only film systems can be further manip-ulated to allow the reproduction of other desired colorpositions. These might include true scene color (perfectmatches to the human visual system) or more novelalternative color such as might be achieved with lesscommon chemical processing techniques on more tradi-tional films—cross-processing (chemically developingreversal camera films in negative film laboratoryprocesses), black-and-white, retained silver (skip-bleachprocessing), etc.

Characteristic Curve Definitions for a Scan-Only Film

Most traditional color motion picture films aredesigned with preferred contrast and color reproductionfor optical or contact printing. In the case of scanning

Designing Camera Origination Films for Scan-Only Applications inTelevision and Digital IntermediateBy David L. Long and Thomas O. Maier

With the growing popularity of digital inter-mediate in the feature film market and thecontinuing development of improvedtelecine technologies for television, thenotion of traditional camera origination filmdesign optimized around direct optical orcontact printing is radically challenged. Forscan-only film image capture systemswhere the camera film will be scannedexclusively to video or data and neverdirectly printed, a very different approachis taken to sensitometric curve design.Rather than focusing on tonal characteris-tics, which yield acceptable and preferredresults for overall contrast, grayscale neu-trality, and shadow and highlight detail in atraditional printing paradigm, the filmcurves must instead be drawn to optimizethe quality of signal provided to the digiti-zation step in the film scanner. Further,standard constructs of color reproductionare also open for modification as the film’sspectral sensitivity and color-enhancingchemistries are changed to better meshwith digital color-processing capabilities.

David L. Long Thomas O. Maier

Long.qxp 3/14/2006 2:37 PM 21

@ SM

PTE All Rights Reserved

67.247.176.132D

ownloaded at SM

PTE Mem

ber on 2012-08-10 from IP

112222 SMPTE MMoottiioonn IImmaaggiinngg Journal, April 2006 • wwwwww..ssmmppttee..oorrgg

applications, however, further manipulation of thefilm’s characteristic response can be made toenhance image quality during digitization.Specifically, the film’s tone transfer function canbe lowered in contrast and reduced in gross fogdensity or minimum density to maximize signalquality for the electro-optical sensor in a typicalscanner. Most film scanning technology is basedon analog capture by way of photomultipliertubes or CCD sensors. When combined with ana-log-to-digital converters and signal processingelectronics they yield a video or digital signal pro-portional to the optical density of the capturedand processed optical film image at any particularpixel measured by the sensor. Whether the film-optical image is described as negative (where thetransfer function of the film is such that repro-duced optical densities are directly proportionalto the level of incident light from the scene) or reversal(where the transfer function of the film is such thatreproduced optical densities are inversely proportionalto the level of incident light from the scene), the opticaldensity representing image content on the film is trans-ferred to the scanner’s sensor. This is done by way of asource light being focused through the emulsion layersof the film and onto the sensor. The sensor, in turn,measures the signal of incident intensity from the atten-uated source light and renders a response level inverse-ly proportional to the measured optical density on thefilm. As is the case with most photon-counting devicessuch as scanner sensors, there is a baseline dark cur-rent noise at low incident light levels (high film densi-ties), which is generally constant, and a photon shot-noise component, generally proportional to the level ofincident light. By designing a film characteristic transferfunction with a lowered contrast and lowered minimumdensity, high reproduction densities, which will translateto low incident photon levels on the sensor, can beavoided. Subsequently, if the scanned film image ismanipulated to a given grading for color and contrast, astarting image with lower raw contrast will provide asuperior electronic signal-to-noise in the final rendering.

By examination of dark current and photon shot noisetheory, it can be concluded that electronic scanningnoise in typical telecines increases as a function ofincreased exposure on the scanning sensor. The signal-

to-noise ratio of the system, however, generallyimproves with increased exposure (Fig. 1) as the totalnoise increases via a typically exponential function withexponent significantly less than 1. Generally, dark cur-rent noise is a constant in linear terms, independent ofthe incident exposure level on the sensor, while shotnoise increases as the square root of exposure.Therefore, by placing as much of the film image as pos-sible on a higher exposure space for the scanner, elec-tronic signal-to-noise performance is greatly improved.Added noise can be minimized in subsequent imageprocessing or color grading.

As further argument to the premise of creating superi-or signal from lower contrast camera films, the conceptof recorded image dynamic range is introduced. Imagescaptured on a traditional color motion picture film that isdesigned for optical or contact printing can exhibit a lossof detail in the highlights of high dynamic range scenesafter an electro-optical scan. This loss is commonly aresult of white “clipping,” wherein the electro-opticalimage path in the scanner system is incapable ofrecording the full range of optical densities reproducedon the piece of film. Additionally, many film scannersuse light sources that are deficient in blue light output.Conventional films have relatively high minimum blueoptical densities resulting from the presence of colorcompensating chemistries and other formulation ele-ments to produce a desired optical print. Thus, they can

DESIGNING CAMERA ORIGINATION FILMS FOR SCAN-ONLY APPLICATIONS IN TELEVISION AND DIGITAL INTERMEDIATE

Figure 1. Typical electro-optical scanner noise profile.

Long.qxp 3/14/2006 2:37 PM Page 122

@ SM


67.247.176.132D

ownloaded at SM

PTE Mem


exhibit excessive blue channel noise in anelectro-optical scanner.

Despite the simplicity of the precedingarguments, the practice of designing filmswith lower contrasts to yield maximumscanner signal-to-noise performance doeshave limits. Specifically, as film contrast islowered, analog-to-digital quantizationsampling comes into play, relative toacceptable contouring artifact levels. Thelower the contrast of the scanned film, themore likely that two consecutive digitalsample levels will represent film densitiesand, further, scene exposures that are sep-arated by more than the human visual con-trast threshold.1 As a consequence, a prac-tical lower-bound to the contrast of a filmcan be defined for scanning applications bythe following equation (1).

!D/!E " (DR/n)/!Ei (1)

where !D is the total range of film reproduction densi-ty over a logarithmic exposure range of !E;

DR is the dynamic range of the scanner in terms offilm density;

n is the number of quantization levels in the scannerA-to-D;

and !Ei is the human visual threshold for exposuredifference (logarithmic again).

In this equation, !D/!E is the intended film transferfunction contrast over a given exposure range. Thepremise of this equation is that any two consecutive dig-itized film densities should differ by an equivalent logexposure range less than the human visual threshold.The equation is extensible to any specific luminancelevel and exposure range over which a definedresponse is desired. Similar relationships can be drawnfor display encoding systems to prevent visual contour-ing in rendered images as well.

Optimum film contrast for scanning is thus boundedby building a characteristic transfer function of minimumcontrast to promote the maximum electronic signal-to-noise profile and dynamic range transfer from scanning.It is also bounded by controlling the contrast reduction toa level that still qualifies under the limits of equation (1),thus preventing contouring artifacts from developingduring the digitization process. Figure 2 presents a sum-

mary of the practical implications of the preceding argu-ments.

When a scanner is set-up for typical transfer, RGBminimum and maximum data points are set, corre-sponding to specific minimum and maximum film densi-ties. However, it should be noted that actual density lim-its can be variable in practice; a function of scene con-tent based on flare considerations within the scanneroptical path. For any given film, these minimum andmaximum densities can be translated back through asensitometric curve to derive the total scene dynamicrange preserved in the scanned signal.

For comparison, a typical sensitometric reproductionis represented by the curve labeled “Vision2.” Further,the scene exposure dynamic range translated to a dif-ferentiable electronic signal is summarized at the bottomof Fig. 2, in the first bar chart. Regions of the scene aredefined as either “shadows” (nonlinear toe of the sensit-ometric curve), “exposure latitude” (linear portion of thesensitometric curve), or “highlights” (nonlinear shoulderof the sensitometric curve). The new sensitometricresponse is proposed and labeled as “Scan-Only Film.”This film system has a lower Dmin, lower contrast, anda longer linear curve section. This enhances both over-all scene dynamic range translated to differentiableelectronic signal and total linear exposure latitude. By



Figure 2. Characteristic response of camera negative films to various scansetups.

Long.qxp 3/14/2006 2:37 PM Page 123

@ SM


67.247.176.132D

ownloaded at SM

PTE Mem


not utilizing the full density dynamic range ofa typical color negative film for production ofan electronic signal, the typical calibrationscheme could be described as “scanner-lim-ited,” with respect to scene reproductiondynamic range. This practice is by no meansuncommon, because density limits are usu-ally set to represent the full shadow to high-light dynamic range of most average scenes.This takes full advantage of digital bit depthin the scanner data path.

For normal scanning applications, it mustalso be recognized that a typical calibrationscheme is not the only available option.Scanners are designed with dynamic rangeprofiles capable of discriminating film densi-ties present in positive, high-contrast repro-ductions. Therefore, the “forbidden zone” ofscanning can easily be opened wider thanthe typical density range represented by alow-contrast color negative film. The blue forbiddenzone in Fig. 2 represents one such calibration scheme.As is evident, it is unlikely that any density reproducedby the Vision2 sensitometric curve will not produce afinal differentiable electronic signal. In this case, the cali-bration scheme could be described more accurately as“film-limited,” relative to scene exposure dynamic range.The bottom bar graphs in Fig. 2 summarize the differ-ence in a film-limited dynamic range scheme versus thescanner-limited scheme for the two film curves shown.By virtue of possessing a longer positive slope beforethe final maximum density plateau, the scan-only filmsystem will still enjoy an advantage over the typicalVision2 reproduction.

In addition to the benefits already outlined here, filmswith lowered contrasts may also be augmented withadditional features including reduced blue minimumdensities (to help with blue-deficient scanner sources),reduced sensitivity to radiation or physical pressuredamage, improved sharpness, and reduced grain.

Extending Usable Dynamic RangeAn important aspect of this system concept is the cre-

ation of a film characteristic transfer function withextended linearity to maximize capture dynamic range.A proposed design comprises a negative color film with

red-sensitive, green-sensitive, and blue-sensitive imag-ing layers. This adheres to the following transfer shapedefinitions:

(1) A photographic speed equivalent to current KodakEI 500 motion picture negative films.

(2) A nonlinear density response versus the logarithmof exposure that is at least 0.7 logE units wide betweena point where the contrast is 25% of the mid-scale con-trast and the onset of the linear mid-scale contrast.

(3) A roughly linear mid-scale contrast that is at least2.0 logE units wide, and (4) a nonlinear contrast roll-offthat is at least 1.8 logE units wide between the end ofthe linear mid-scale contrast region and the point wherethe contrast has dropped to 25% of the mid-scale level(Fig 3).

The upper-scale specification is especially significantas it addresses the problem of faster contrast roll-off inconventional motion picture films. By defining a highercontrast farther into the upper end of recorded dynamicrange, the proposed design offers significant improve-ment in the quality of rendered highlight informationupon electro-optical scanning. Rather than being generi-cally defined by standard ANSI statusM optical densityspecifications, the preceding design features are alsodefined exclusively to scanner density specifications,calculated as the integral product of film dye spec-



Figure 3. Basic linearity definitions for scan-only film system.

Long.qxp 3/14/2006 2:37 PM Page 124

@ SM


67.247.176.132D

ownloaded at SM

PTE Mem


trophotometry and electro-optical system spectralresponse.

A final important point to be made regarding this con-cept is that the absolute level of contrast is not impor-tant, relative to the creation of a first-generation aesthet-ically pleasing image. Instead, it is defined exclusivelyagainst the creation of superior electronic data uponscanning. Further, color reproduction techniques incor-porating masking, inhibition, and contamination strate-gies can be widely modified from the general hue, satu-ration, and neutrality control objectives representative ofnormal color negative films. Typical color negative print-ing or scanning steps would lead to very flat and poten-tially non-neutral image reproduction using the proposedfilm curve, if subsequent contrast enhancement werenot utilized. However, this film is intended to be usedexclusively in combination with digital image processing,which yields a pleasing final look. Thus, the aesthetic ofthe raw translated image is not a factor for this design.Instead, the proposed tonescale adjustments here areintended to enhance the quality of scan data, relative toscene dynamic range and electronic noise.

Impact of Color ChemistryBeyond tonescale design, further optimization of the

scan-only film is accomplished by removing color chem-istry components such as masking couplers and inhibi-tion couplers. These are designed to impart specific pre-ferred color and tone reproduction characteristics in tra-ditional direct print image capture and distribution sys-tems. With these elements removed, the film systemcan be designed to maximize light-capture efficiencyand minimize image noise or graininess. Color-controlcomponents built into traditional image capture films candegrade image quality by reducing the efficiency of lightcapture and subsequent amplification during chemicalprocessing. By utilizing digitization and manipulationsteps subsequent to the capture of the image by thefilm, any desired image look can be introduced into thefinal display and distribution chain.

Practical Applications for a Scan-Only FilmDesign

One of the primary benefits of a scan-only film systemis that it permits the combination of the superior lightcapture capabilities of a photographic film, with the

enhanced flexibility and efficiency of digital post-produc-tion techniques. By trading off the requirement of havingan image origination film stock produce an optical imageembodying a preferred look when directly printed orscanned/transferred to video, a single scan-only film canbe created and optimized to capture additional informa-tion from a scene. Additional scene information allowsvarious existing camera negative film looks to beapplied with a processor, employing algorithms that rep-resent select film color and tone characteristics. Multiplefilm origination stocks are not needed to achieve thevarious film looks exhibited by conventional originationstocks. Looks are primarily achieved with photoscienceimage processing algorithms rather than with the specif-ic emulsion chemistries that provide each film with itsown unique look. This concept is easily extendable toeither television or digital intermediate applications.

Alternate Color Reproduction ConceptsEnabled by Scan-Only Systems

Being able to render the tone and color qualities ofany past or present color film stock from a single startingfilm image offers significant benefits to creative andtechnical content producers. However, it is clearly notthe only potential advantage of a scan-only film.Marrying a scan-only film with more advanced process-ing algorithms that simulate optical or chemical effects,otherwise cost- or time-prohibitive (such as cross-pro-cessing, skip bleach processing, etc.), allows users todo much more with their images in far less time. Further,alternative color reproduction paradigms could also beexplored. One such application which deserves specialattention focuses on designing a film that offers com-pletely accurate color reproduction across the full gamutof visible colors.

A significant difference between traditional color origi-nation films and the human visual system is the specificrelative spectral sensitivities each has to regions of red,green, and blue light. Figure 4 illustrates the difference.In Fig. 4, the spectral sensitivities have been normalizedto equal area to provide equivalent color balance. Inpractice, the film will be exposed through some typicallens on the camera, and the short wavelength sensitivityof the film will be removed by the UV absorption of thelens. This leaves the film and human eye blue sensitivi-ties reasonably close. Film and human eye green sensi-



Long.qxp 3/14/2006 2:37 PM Page 125

@ SM


67.247.176.132D

ownloaded at SM

PTE Mem


tivities peak at about the same wavelength.However, the film sensitivity possesses a muchhigher peak, because the film sensitivity is muchnarrower than the human eye sensitivity. Thetwo sensitivities differ the most in the red region.The human eye red sensitivity overlaps thegreen sensitivity significantly, whereas the filmred sensitivity is shifted to longer wavelengthswith considerably less overlap.

There are several reasons why traditionalcolor origination film sensitivities differ from thehuman eye. One can be found in the image pro-cessing done in the human brain versus thatdone chemically in the film. The currently accept-ed color processing model for the brain is theOpponent-Color theory.2 This model involvesthree channels of color, but not the red, green,and blue channels that are captured by thecones in the retina. Instead, the signals from thecones are combined to produce three signalsrepresenting a function of (1) red + green + blue(2) red-green and (3) red + green-blue. Thesethree signals simplify to define a total signal, a redness-greenness signal, and a yellowness-blueness signal. Inthe classic chemical processing model found withincolor film, it would be impossible to provide for thesetransforms of the original red, green, and blue capturedsignals. Additionally, because color origination film isprinted onto a color print film, the captured red, green,and blue signals must be transferred to complementaryprint film dyes, not color difference signals.

Another reason for the difference between color filmand the eye can be found in the arrangement of the sen-sors in the two different systems. In the human eye, thecones are packed adjacent to each other and each conehas a red, green, or blue sensitivity. In film, the layersare stacked on top of each other. To understand theconsequence of these different arrangements, take forexample a 580 nm photon. In the eye, this photon couldhit either a red-sensitive cone or a green-sensitive coneand produce a signal from that cone. The probability ofproducing a red or a green signal depends on the rela-tive number of red and green cones and the relativesensitivities of the red and green cones as shown in Fig.4. There are many photons at normal viewing lumi-nances; therefore the red- and green-sensitive cones

will absorb enough photons to allow a person to per-ceive the correct color. However, in film, the layers arestacked on top of each other with the green-sensitivelayer above the red-sensitive layer. Therefore, any pho-ton absorbed in the green-sensitive layer will not reachthe red-sensitive layer. So, if the 580 nm photon isabsorbed as it passes through the green-sensitive layer,there is zero probability that it will be absorbed in thered-sensitive layer. This forces a film design that mini-mizes the overlap of the red, green, and blue sensitivi-ties. However, it is the overlap in sensitivities that allowsan observer to distinguish between two wavelengths oflight that differ by only 1 or 2 nm in the human visualsystem.

The film system cannot match the human eye spec-tral sensitivities or perform the processing that is per-formed in the brain. Color film historically produces col-ors that are not exactly the same as those perceived bya human observer in the original scene. In fact, althoughfilm colors are reasonably accurate, they have actuallybeen optimized historically to offer the most pleasingcolor rendition, as determined by a large number ofviewers averaging reproductions over all possiblescenes. This is traditionally accomplished with a care-



Figure 4. A comparison of human and a daylight film spectral sensitivity.

Long.qxp 3/14/2006 2:37 PM Page 126

@ SM


67.247.176.132D

ownloaded at SM

PTE Mem


ful engineering of the spectral sensitivities and colorprocessing chemistries found in modern color films todevelop a desired full-system color reproduction.

If complete color accuracy in film reproduction isdesired, it is theoretically possible to produce the sameor very nearly the same colors on film as the humanobserver sees. This is accomplished by replacing theclassic film spectral sensitivities with another set that isa linear combination of the human spectral sensitivities.To recover the same information as was captured by thehuman eye, the reverse of the specific linear transformchosen needs to be applied to the film captured imagein processing. Again, this is impossible in the traditionalchemical technologies available with film. With digitalimage processing following a scan step, however, thetransformation becomes trivial.

A color film designed in this manner would not neces-sarily offer classically preferred image reproduction if itwere to be printed directly onto a color print film.Further, an altered spectral sensitivity film, althoughcapable of producing more accurate color reproduction,will not be able to produce the color reproduction ofexisting films. Such is the disadvantage of designing afilm for a very specific application. Once spectral sceneinformation is integrated and recorded with a set of sen-sitivities not linearly convertible to another known posi-tion, no amount of image processing will bring the colorreproduction back to the same point.

The image processing needed to produce images thatmatch the colorimetry of the original scene is not diffi-cult. If a linear combination of the human eye sensitivi-ties was used in the film, the film densities would needto be converted to film exposures using the characteris-tic film curve. Following this, a 3 x 3 matrix should returnthe exposures to values representative of those cap-tured by the human eye. Because the color matchingfunctions used to calculate the CIE XYZ tristimulus val-ues are a linear combination of the human eye sensitivi-ties, another 3 x 3 matrix would provide these. With tris-timulus values known, several recognized scene-refer-enced output algorithms could be used to generateappropriate data for creating a reproduction colorimetri-cally equivalent to the original scene. The final resultwould be a film system that offers truly accurate colorreproduction as an alternative to more classic prefer-ence-based color film systems.

ConclusionWith the advent of newer and better scanning tech-

nologies for motion picture film, a new paradigm for filmdesign is required. The elimination of the classic require-ments of direct optical printing and chemically derivedcolor reproduction open the door to allow film to be abetter sensor of light and a better starting point for high-quality images. By taking advantage of the power andflexibility of digital image processing, the imaging bene-fits of silver halide capture can be augmented to pro-duce truly superior digital data for any present or futureimaging application.

References1. Peter G. J. Barten, “Contrast Sensitivity of the Human Eye

and Its Effects on Image Quality,” SPIE—The InternationalSociety for Optical Engineering, 1999.

2. R.W.G. Hunt, The Reproduction of Colour, 5th ed.,Fountain Press: England, 1995.


Presented at the 147th SMPTE Technical Conference and Exhibition in NewYork City, November 9-12, 2005. Copyright © 2006 by SMPTE.


David Long joined Eastman Kodak Co. in 1997 as amember of the Manufacturing Research and EngineeringOrganization. In 1998 he moved into his current role asproduct development engineer and hybrid imaging scientistwith the Entertainment Imaging Division. His primaryresponsibilities with Kodak include new product develop-ment and image science and systems integration for themotion picture group, focusing on film and hybrid imagingproducts. Long joined Kodak after graduating from theUniversity of Texas at Austin with a BS in chemical engi-neering and from the University of Rochester where heearned a MS in materials science.

Tom Maier is a research fellow in the EntertainmentImaging Business Unit of Kodak. During his 33-year careerat Kodak, Maier has written computer programs to modelfilm and digital imaging systems, developed methods tocharacterize and optimize the color quality of these imag-ing systems, and has run psychometric experiments toconfirm the image-quality of his computer optimizations.Maier has served on a number of the CIE committeesrelated to color, as well as several SMPTE DC28Standards Committees for Digital Projection Systems. Heearned a BS degree in chemical engineering from MIT anda PhD degree in chemistry from the University of Illinois.He is a member of SMPTE.

THE AUTHORS

Long.qxp 3/14/2006 2:37 PM Page 127

@ SM


67.247.176.132D

ownloaded at SM

PTE Mem


designing camera origination films for scan-only

Documents