Digital Color Management Basics. An Introduc6on for Mo6on Pictures

Transcription

1 Digital Color Management Basics An Introduc6on for Mo6on Pictures

2 Reference Reading Review Hunt, Chapter 11, Objec6ves in Colour Reproduc6on 2

3 Defini6on of Color Management Every successful color-imaging system employs one or more means for controlling and adjusting color information throughout the system. That is what is meant by color management. - Digital Color Management by Ed Giorgianni 3

4 Mo6on Imaging Color Systems Capture Process Output Would you rather design: - nin x nout systems - nin + nout systems 4

5 Color- Encoding Capture Process Output Convert the light at each picture element into a number. A function of device design. Manage/Convert the device input numbers to a common encoding. Prepare/Convert numbers from encoding for output device. Convert the number associated with each picture element into light. A function of device design. In generic color-encoding each picture element requires a: RED Code Value GREEN Code Value BLUE Code Value 5

6 Mo6on Imaging Systems that are Color Managed Conven6onal op6cal film systems Nega6ve to print Nega6ve to intermediate(s) to print Video Systems From video capture From telecine scan of film Digital Data Systems Digital Intermediate Data capture CGI Video capture Management can be analog (as we ve already studied with film design and Jones diagramming) or digital 6

7 Imaging System Paradigms Type A: Input Driven Output image colors = Input image colors Photocopy by design: Example Input: Camera Nega6ve Example Output: Duplicate Nega6ve Mo6on Picture Implica6on: The Duplicate Nega6ve is an exact copy of the Camera Nega6ve (*at least as far as the print film is concerned*) 7

8 Imaging System Paradigms Type B: Encoding Driven All output images = render of encoded colors Encoded digital Code Values are resident in a file Example Input: CGI on a monitor Example Output: rendered image on color print The CGI ar6st is not interested in color problems between crea6on and exhibi6on; he just wants his colors faithfully represented Mo6on Picture Implica6on: The monitor image and the print film image match the color represented by the code values in the file 8

9 Imaging System Paradigms Type C: Output Driven The output image colors will be op6mized by the output device Example: TVs on the wall at Best Buy - Most saturated monitor image - Sport mode, Vivid mode, Movie mode???? Mo6on Picture Implica6on: Electronic image does not match the printed image as each is differently able to realize the goal 9

10 Profile Connec6on Space (PCS) Input PCS A common language based Output in colorimetric standards of an appropriate type 10

11 Don t Forget Color Appearance! The PCS should respect color appearance phenomena and viewing condition influences as well - Flare - Chromatic adaptation - Dark-surround effects - Luminance effects - Etc 11

12 Image State Diagram Image States in which mo6on imaging colors can be encoded: Scene Colorimetry/Appearance Encoding (exact colorimetric or appearance reproduc6on: think ideal video cameras and produc6on telecines) Prin6ng Density Encoding (view of the world as seen by a nega6ve film a preferred distor6on of reality as manifested in capture film) Rendered Appearance Encoding (image manipulated in op6cal prin6ng or in video or data to produce a desired output In addi6on to print film density, standardized video (709) & Digital Cinema color encoding are based on this paradigm 12

13 Mo6on Picture Image State Diagram Scene Encoding Color and Tone Render Rendered Appearance Encoding Negative Film Model Print Model Printing Density Encoding Translation between spaces usually infers intentional color manipulation, both artistic and technical! 13

14 Mo6on Picture Image State Diagram Video Camera Digital Camera Film: Reversal Print Film: Reversal Color Negative B&W Scan-only Restoration Intermediate Film Scanner Telecine Scene Encoding Color and Tone Render Rendered Appearance Encoding CGI Negative Film Model Creative Effects Image Correction Print Model Film Writer Print Digital Projector Film: Color Negative B&W Scan-only Restoration Intermediate Film Scanner Printing Density Encoding Film Writer Intermediate Video Display Optically Print Print 14

15 Film System Color Management Film system color management is based on: The manufacture of film within specified tolerances Spectral sensi6vity Sensitometry Interimage Dye spectra The manufacture of chemical process solu6ons within specified tolerances Control strips that allow the users to characterize the chemical and op6cal post- produc6on processes Procedures to put the processes in calibra6on based on the results from the control strips 15

16 Digital System Color Management There are mul6ple digital systems that now touch mo6on imaging content and enforcing the common interchange enjoyed with film systems is cri6cal Digital system color management is based on: Analyzing the needs of the imaging system Designing a color management system with a chosen encoding metric Characterizing each element Calibra6ng (controlling) each element 16

17 Digital Color Management for Digital Intermediate & Data DataCam/Scanning & Recording (and Manipula6on in Between)

18 Data and Video are Distinctly Different Film Path Film Capture Copy and Duplicate Theatre Telecine Scan Record Digital Postproduction Manipulate, Effects and CG Lossy Compress Digital Cinema Data Capture V to D DDR D to V DDR Video Capture *Decomp Video Postproduction Lossy Compress DVD TV Electronic and Video Paths 18

19 Data and Video We ve described data vs video in a Dailies workflow We ve described data vs video in electronic capture paradigms We ve described video display chain color and tone reproduc6on characteris6cs Color- matching func6ons, display primaries, matrices, and the match to scene colorimetry We ve described film s place in a video workflow Produc6on vs projec6on telecines and manual aesthe6c tweaks 19

20 We ve Outlined the Technical Differences Digital Video Postproduction (DV) Data (DI & Datacam) Signal Precision In a stream like water in a pipe MOV, MXF Low precision Generally some form of 8 to 12 bit linear with gamma In computer files generally CINEON, DPX, RAW, TIFF High precision - 10 bit log or 12-16bit linear going to 32 bit FP Speed Real time Non real time but getting better Resolution Color subsampled Up to 1920 pixels/line Almost always 2048 to 6224 pixels/line and beyond Never (except in datacams) Compression Always Never (except in datacams) Dynamic Range Supports 10 stops gamma encoded Supports 20 stops (except in datacams) 20

21 Let s Add One More!!! Digital Video Postproduction (DV) Data (DI & Datacam) Signal Precision In a stream like water in a pipe MOV, MXF Low precision Generally some form of 8 to 12 bit linear with gamma In computer files generally CINEON, DPX, RAW, TIFF High precision - 10 bit log going to 32 bit FP Speed Real time Non real time but getting better Resolution Color subsampled Up to 1920 pixels/line Almost always 2048 to 6224 pixels/line and beyond Never (except in datacams) Compression Always Never (except in datacams) Dynamic Range Supports 10 stops gamma encoded Supports 20 stops (except in datacams) Standardized? Absolutely! Sort of! 21

22 Data and Video We ve hinted at data s place in a digital intermediate environment for theatrical exhibi6on We ve hinted at the Reference Prin6ng Density paradigm for data colorspaces Logarithm of print exposure We ve stated professional theatrical exhibi6on is a data world, not a video world 22

23 But the lines are blurring!! Produc6ons want one look for television and theatrical distribu6on Are they sure! They also want to get it from one scan Video cameras can be used for DI Video transfers of film can be used for DI Think about your own SOFA films Did you make them for television or the theater? What do you want them to look like? 23

24 And so let s talk Data but from the perspec6ve of the film system

25 Conven6onal Duplica6on Color Original Negative Work Print Color Master Positive Color Duplicate Negative Color Release Print 25

26 Scanning and Recording Color Original Negative Work Print Film Scanner Color Master Positive Color Duplicate Negative Electronic Intermediate Film Recorder Color Release Print Duplicate Negative 26

27 Or Shown Another Way Digital Mastering (Data) Digital Intermediate Film Production (Data) Film Recording Scanning (Data) Data Camera (Data) 27

28 The Resolu6on Game The industry is slowly migra6ng from a 2K philosophy to 4K for capture It was at one 6me that 4K was only considered for serious effects work Higher end cameras and scanners are all marketed as 4K+ capable This puts tremendous strain on hardware and storage infrastructure Even distribu6on must decide what to do with elevated data rates DCI does make room for a 4K projec6on format though the vast majority of currently installed systems are barely bemer than HD What is the business model for upgrading? 28

29 Film: How Do We Go From an Image to Numbers That Aren t Video? Opal Glass Optics Film (emulsion side down) Fiber Optic Bundles It s just the physics of light quantified But in density, not exposure!!! Log radiometry, not linear R G B Photomultiplier Tubes with Filters 29

30 Clearing Up Linear vs Log Thanks to the influence of a large number of engineers who don t really know any bemer, the terms linear and log get thrown around very loosely in color space descrip6ons The power of marke6ng and branding The real reason for these differences? Video engineers who don t understand integral density or film sensitometry describe anything filmic as log Film engineers who are accustomed to log- log characteris6c curves describe anything videoish as linear We re all taught that one or the other is clearly superior in every situa6on when in fact they each have their place in device physics and image processing Color management can ul6mately relate the two spaces! 30

31 Linear What would be most correct for a linear signal: a color space representa6on directly propor6onal to linear exposure or transmimance for an imaging system having a defined spectral response and u6lizing a given balance criterion Tris6mulus!!! But ooen used also to describe these situa6ons in mo6on picture: A video signal derived from a linear camera exposure that has undergone non- linear color or tone manipula6on The linear sensor transmimance of a telecine or film scanner (which includes film tone characteris6cs as well) A video signal complete with gamma compensa6on applied The an6log of a signal someone has described as log Others I m probably leaving out 31

32 Log A log signal is one representa6ve of op6cal density or analy6cal dye amount as jus6fied by Beer- Lambert physics It is perfect for any imaging workflow where the final deliverable is a piece of film Recall nega6ve film density (- log 10 T) is the loge to the print film Other possible uses: Any digital film scan signal, regardless of the quality of calibra6on or signal processing employed The logging of a signal someone has described as linear à par6cularly popular in the digital cinema cameras used today and a necessary step to make them fit into the contemporary digital intermediate workflow Panalog S- log (Sony) Etc 32

33 What Should You Do?? Be smarter than the manufacturer s spec sheets and use common sense color science to determine what you re really working with These classes give you all the tools you need to interpret any system you encounter Remember the rules of device- dependent color: a 3- channel signal is only per6nent to color appearance in the proper context Once you acknowledge the physics of an imaging situa6on, you then use the math to get what you want There s nothing wrong with other s claims but you should always work in data you fully understand and can fully manipulate 33

34 Prin6ng Density Metric of Choice for Scanning Film to a Log Signal 0.40 Typical Well-Designed Film Scanner Response Scanner Response Print System Response Film density is log Exposure to the next film is log It s a log-log world Wavelength (nm) Resp(") = Illum(") # SpecSens(") 34

35 Integral Density Review Full mathema6cal representa6on of integral density For prin6ng density, the source illuminant is the printer and the detector is print film T = 10 "D = % % SPD(#) out $# SPD(#) in $# = % T(#)S(#)SS(#)$# % S(#)SS(#)$# Where S(λ) is the spectral power distribu6on of the source illuminant and SS(λ) is the spectral response of the detector 35

36 The Real Reason for Reference Prin6ng Density The vast majority of the world s theatres are s6ll projec6ng film This means that any mastering of color is usually done in a filmic context Film recorders are responsible for taking manipulated reference prin6ng density and making it real prin6ng density for mass duplica6on via an intermediate film 36

37 Data Color Depth With the Cineon Digital Film system in the 1990s, Kodak established the current industry norm for color encoding 10bit RGB color Prin6ng density encoding (logarithmic) 0.002PD per CV 37

38 Proper6es of the Nega6ve Typical Negative dynamic range easily fits into the printing density span afforded by the Cineon definition (at least 15 stops of scene exposure) bit Density Code Values D stops Density per Code Value Relative Log Exposure = Printing Density 38

39 Tonal Range in a Typical Scene D 1023 = Density per Code Value bit Density Code Values Stops 3 Stops Printing Density % gray Relative Log Exposure Don t follow this CV-to-PD conversion for the negative film too closely à other conventions are employed typically 39

40 Scanner Opera6on Balanced on film D- min (generic balance is ) One code value = prin6ng density Recall a typical printer light is loge to the print film 12.5 CVs per printer light Film D- min CV > 0 to allow for grain fluctua6ons Where possible, normal 18% density should be balanced to 445 CV (doesn t always match with guideline above) Requires some knowledge of film stocks and some calcula6ons frameline Dmin samples are typical opera6ons balance cue 40

41 Code Value Conven6ons Why place the scene 18% gray at on the scanner? this translates to 0.89 PD above Dmin on the duplicate nega6ve film to be recorded (we ll talk about how recorders are calibrated in a bit) Taking typical intermediate film Dmin of 0.05, 0.55, 0.6 and adding 0.89, we get very close to Kodak recommenda6ons for duplicate nega6ve LAD placement (1.00,1.45, 1.55)* *yes, I m mixing StM and PD just an approximation 41

42 Code Value Conven6ons Why place the nega6ve Dmin at a balance on the scanner? As it turns out, the average density range between Dmin and 18% gray on a properly exposed nega6ve film is 0.7 With 445 chosen to accommodate duplicate nega6ve placement, 95 calculates out for Dmin Are either conven6on solidly followed? No! but it s close 42

43 Kodak Vision2 500T Prin6ng Density A Digital Dilemma: Recall that Color Negative RGB contrasts are not equal in printing density Sure, you can force the 18% gray to R=G=B CV at 445, but that equality quickly disappears up and down the scale Density PD_2383 Red PD_2383 Green PD_2383 Blue This is perfectly OK for film data but it causes graphic artists accustomed to digital images with R=G=B being neutral headaches! loge (log lux- sec) 43

44 Scanner Calibra6on Colors SD [3x3] PD SDr SDg SDb PDr PDg PDb [3x3] = M11 M12 M13 M21 M22 M23 M31 M32 M33 & PDr = M11 * SDr + M12 * SDg + M13 * SDb PDg = M21 * SDr + M22 * SDg + M23 * SDb PDb = M31 * SDr + M32 * SDg + M33 * SDb 44

45 The DPX File Mo6on data files are most typically stored in the Digital Moving Picture Exchange (DPX) format Standardized in 1993 as SMPTE 268M Resolu6on independent Bit- depth independent All file amributes iden6fied in a binary header Raster essence packing 45

46 Image Manipula6on Workflow The DI Color Grading Suite is ooen referred to as a Virtual Telecine Scanned data files are fed from a storage server to color correc6on equipment which provides an HD preview at 24fps or appropriately transformed data rate Preview may be a typical video monitor or a digital projector (either must be calibrated to final film- out colorspace) Full resolu6on manipulated data files are wrimen back to storage Scanner Storage Manipulate Visualize Scan 46

47 Image Manipula6on Workflow As the scanned data files are already being converted to a video signal for preview, it is a trivial step to record the video signal out for typical dailies duplica6on This is the onus for data dailies The real trick comes in conver6ng scanned data files encoded in prin6ng density to video primaries that emulate recorded/printed film outs Scanner Storage Manipulate Visualize Scan Color Management! we ll hit this topic in coming lectures DVD 47

48 Image Manipula6on Workflow Modern color grading is increasingly sooware- based for scalability Bandwidth problems can be prevalent with intensive processing DI suites are also using full size theatrical projec6on to re- create the environment of film dailies It s big business and it s all about (the) image Scanner Storage Manipulate Visualize Scan 48

49 Image Manipula6on ASC CDL As we learned in studying film- to- video applica6ons, the industry has grown up with an expecta6on that electronic color images can be managed with fairly powerful curve- shaping tools Lio, gamma, gain, secondaries, satura6on, mames, etc. But in tradi6onal prin6ng applica6ons, the no6on of printer lights always provided an extremely powerful communica6on tool for the ul6mate look Universally interpretable and understood Universally simple Universal color reproduc6on Electronic systems are notoriously poor in achieving a standard set of controls Everyone s controls are different and worse yet, everyone s physical summary of the amount of any control are different The American Society of Cinematographers, frustrated by this new reality in the digital world, has proposed a solu6on 49

50 ASC CDL The ASC s Color Decision List presents 9 numbers per color decision in a keykode/6mecode- synced edit file that accompanies raw footage Slope, power, and offset for R, G, and B channels look is never baked in un6l ready XML file implementa6on is universal Though a color scien6st may admonish the prac6cality of using the classic lio, gamma, gain approach on any data, linear or log, the importance of the ASC CDL is as a communica6on tool first Future implementa6ons may include satura6on matrices and appropriate linear- to- log conversions 50

51 ASC CDL Input code values are normalized on a 0-1 scale based on signal bit- depth (typically 10 or 12) S = input" slope 0 slope O = S + offset - offset though -1 offset 1 is practical range output = [ clamp 0 "1 (O)] power 0 power output 0 "1 = [ clamp 0 "1 (slope# input 0 "1 + offset) ] power 51

52 ASC CDL - Advanced Many in the ASC demanded 2 addi6onal func6onali6es be integrated into the ASC color corrector Lio changing black levels while keeping whites constant slope = lift & offset = 1" lift Original slope and offset manipulations are allowed and must be properly combined with these results from a lift operation to fit into the ASC s equation Printer Lights a true offset of bit CVs per light; can only be implemented trivially in ASC s equa6on when power = 1 output 0 "1 = [ clamp 0 "1 (slope# input 0 "1 + offset) ] power + PL output 0 "1 = [ clamp 0 "1 (slope# input 0 "1 + (offset + PL _adj))] power 52

53 A Full Rendering: The Print Look Digital Intermediate color work would be fairly useless if the ar6sts couldn t get a feel for exactly how the PD- encoded images would look on projected print film In color management terms, this requires color science algorithms to render forward the appearance of mo6on picture print film Print Film Emula6on! The fun part is that you ve already started to understand these algorithms when you built the Jones diagrams! 53

54 Display/Preview Data File Reference PD Film Recorder Printer R B Dim Surround Additive System M G C Y Dark Surround Subtractive System Matching the additive display to the eventual projected film print involves some complex math We ll be covering this in detail in coming lectures 54

55 Geyng Back Out to Film Data File Film Recorder Laser, Intermediate CRT, Camera Neg Printer, Print Film Printer, Print Film 55

56 Calibra6on Objec6ve Digital CVs are encoded prin6ng density Duplicate Nega6ves should be thought of as yielding prin6ng density responses as well The natural calibra6on goal for a recorder is to produce prin6ng density above Dmin equivalent to input CV*

57 Nega6ve Films Used for Recording Status M Density EASTMAN EXR 50D Film 5245 and KODAK VISION Color Intermediate Film 5242 Sensitometry Red 45-Green 45-Blue 42-Red 42-Green 42-Blue Step 57

58 Introducing Digital LAD Add this digital file to a data sequence to be recorded out The 445 patch can check recorder +processor calibration (0.89PD above Dmin) But it is also a printing guide for generating properly colorbalanced prints 58