Näytetään tekstit, joissa on tunniste Xrite. Näytä kaikki tekstit
Näytetään tekstit, joissa on tunniste Xrite. Näytä kaikki tekstit

21.11.2016

Emulating analog film digitally Part 2 - Emulation process





There’s lot to choose from both free and commercial tools on the market who claims to mimic the look and feel of analog film. I have tested couple of them and the results are not too reassuring.


I tested one product and its emulation of Fuji Velvia 100. Picture on the left is a scan of real Fuji Velvia 100, second one is exactly the same scene shot with Sony A7 with default settings. The third one is interpretation of Velvia 100 from a commercial software. I don't know how their emulation was derived, but it´s quite clear they've never seen a picture of Fuji Velvia before. interpretation is not even close. I was able to do just the same by only editing the contrast in the RGB-curve

                     Velvia 100                                                      Sony A7                                                    Velvia 100 Emulation





At this stage it´s quite clear if I want the look of film digitally I have to do it myself, I can´t afford to buy every piece of software and test them separately if someone happens to get it right. For the process of film emulation I need the following tools:


  • Proper digital camera
  • Old analog SLR (mine is Olympus OM 20) with a lens without color cast, some older lenses develop a color cast while ageing.
  • Film you want to emulate. Shoot the same scene with digital and analog. Different scenes, lighting, high contrast, low contrast, color targets etc.
  • Color checker and many shot of it 
  • possibility to scan the film with unaltered colors. I wrote about this in my previous post.
  • Software to tweak colors. I used Adobe Photoshop and Camera Raw (Lightroom) and my recipe is based on these tools. 


The process



Proper emulation of the film involves emulation of:


  • Dynamic latitude and contrast (tone curve)
  • color response (RGB curves, white balance and color balance)
  • Details (grain and sharpness)
  • Scratches and dust (optionally) 


Start with a pictures taken both with film and digital camera from the color target in a daylight illumination or flash. Both pictures shall have the exact same exposure. Read the color values in all patches from the analog picture, It´s easy with Photoshop or Camera Raw. You need both RGB-values and Lab-values. Both pictures must be in the same color space; sRGB, Adobe RGB or ProPhoto. It doesn't matter which color space, but the color space must always be identical in all pictures. RGB values are different in different color spaces. Film is hardcoded to the daylight white balance (5500 K), so the digital picture must be adjusted according to that. We're trying to emulate the look of the film, not the scene. 

Target Reference - Velvia 100

Unedited picture from the same scene with digital camera



Read the Lab-values from the six gray patches in the digital picture. Adjust the tone of RGB channel in Camera Raw so that the L: in the Lab channel is identical with the analog picture. L: channel does not contain any color information only lightness. You should now have some kind of s-shaped tone curve. Then change the readings from Lab to RGB and adjust all the three channels individually so that all RGB channel values are identical to the analog picture. Much of the color information in analog film is "hidden" in the gray patches – film is never completely white balanced. When this step is completed, your digital picture should already be very close to the target. Color response should now be almost accurate and only some remaining colors may still need some extra tweaking, but you can leave it here if you want because curves responds to the light in a “right” way and won't twist colors in different lighting situations. 

Digital picture after the modification of RGB-curves



Next step is tweak with HSL sliders for the remaining colors to match. Most important is the hue. Hue tells what color the color is. This is a time consuming process and it´s impossible to get a 100 % match. More is not better here because too many/extreme corrections will only twist the colors and the preset will be less universal. Try to keep individual color modifications to the minimum. In theory you should be able to mimic the look of the film only by adjusting the RGB curves. Film is different from digital camera in way that the color response is not linear i.e. the hue of the individual color will change along with exposure.


I also tried to make camera specific DNG profiles with DNG profile editor but it didn’t gain any better results.


Then compare analog/digital shots from different scenes and lighting and tweak even more. This is rather an artistic process than scientific. But don't overdo it, every scene will not be picture-perfect no matter how you tweak it.


When the colors are ok you can add the film grain. In Lightoom/Camera Raw just inspect the grain in the analog picture and add grain so that the amount, size and roughness will match. Velvia 100 has the values: Amount: 12 Size: 42 Roughness: 76. If you want the perfection, you can copy the grain from a neutral area in the analog picture and blend it in a layer in Photoshop, but I don’t think anyone will see any difference and the process is time consuming. Graininess in film increases as density increases and film grain is not monochromatic. Too much artificial grain will tell it´s “photoshopped”.


My emulation of Velvia came across better than I expected and it works well in wide variety of lighting situations. It's no trick to fix one photo to match another, but automatizing the process is. Of course not every color is spot on because films response to light is not linear, but the feeling of Velvia shines through and it’s hard to tell difference especially if you don't have a reference to compare with.


It´s a mystery to me why these commercial film emulating products doesn´t perform any better than they do. Is it lack of skills or improper scanning or what?


Here are some examples I compared my interpretation of digital film with the original and a well known company who provides presets to mimic film:

Velvia                                Straight out of camera                   My version                                   Commercial version

I also did preset for Kodak Ektar negative film:


Kodak Ektar                                                My version                                                   Commercial version

In the last picture pair one is real Velvia and the other one is my simulation of it. It´s hard to tell the difference:


This is Velvia                              Or this?

27.10.2016

Should I calibrate my camera?






All cameras produce slightly different colors and if shooting RAW, then different RAW developing software will interpret colors in a slightly different way. Virtually all camera sensors today are somewhat similar CMOS-type with Bayer RGB-filters which will guarantee that the colors from different cameras are very close to each other. There are some differences, but they’re very subtle compared to differences between films. If shooting JPEG, every camera manufacturer has their own recipe how they express colors. You may also choose between the color style e.g. Portrait, Landscape, Neutral etc.

There’s an ever going debate between gear holics on which camera has the best color; Canon or Nikon? Olympus produces best colors they say, Sonys colors are bland. Capture One has better colors compared to Lightroom and so on.

I shoot a lot of food photos where accurate color reproduction is to be expected. My main cameras are Sony A7 and Canon EOS 5D Mark II. My subjective judgment is that Canon has more pleasing colors to the eye compared to Sony, both in JPEG and through Lightrooms mangle. Sony’s colors are harsh, where Canon has some kind of pleasant warmth in its colors. Here I must stress, that accurate and pleasing colors have nothing to do with each other. Accuracy is an objective term and pleasing is most subjective. You can´t measure how pleasing colors are, but you can measure accuracy which I will do here.



Calibration


If you shoot raw, you can calibrate your camera for accuracy or even get two cameras of different brands to produce colors alike. Because I use Adobe Lightroom, I will discuss about DNG profiles. Other raw converter will read ICC profiles. Concepts are the same in both.


Camera JPEG


When you shoot JPEG or RAW converted with camera manufacturers own RAW-converter you will get colors the manufacturer intended them to be in the first place. Because the manufacturer knows its camera best, you could expect best results here. If the colors are off, it’s not by accident.


Adobe Standard profile


When you import your RAW-photos into Lightroom, photos will be converted into the Adobe Standard profile. The RAW file has to undergo demosaicing and other adjustments before you can call it a photograph. Lightroom must recognize the RAW-format of your camera, otherwise it can’t read them at all.


Own custom profiles


If you’re not happy with the Adobes interpretation of colors, you can make your own profiles. There are several ways to create a desired profile and the methodology is same for them all: Shoot a standard illuminated color calibration target and compare the results to the reference values and you will get a calibrated camera with authentic color reproduction, at least in theory. I tested to create profiles with Adobe own DNG Editor, Xrite editor and QPCard. Adobe DNG Editor and XRite both uses the Color Checker as reference. QPCard has their own reference target which they claim to be more suitable for camera profiling than Color Checker thus giving better results.


Color errors


I did my test in sRGB color space which is the narrowest, but virtually every photograph will be converted to sRGB in the end, so this is justified and the colors are comparable in the web browser.

Color consists of hue, saturation and lightness. Hue is the most important, because it will tell what color the color is. Saturation indicates the degree to which the hue differs from a neutral gray. Usually cameras will exaggerate the saturation. Luminosity is the brightness of the color and will change with brightness and contrast and can easily be corrected in post. If the hues are wrong, the colors are wrong.
Methodology

I created my own (single illuminant) camera profiles as thorough I was able to. Then I shot a color checker target in standard illuminant (D50, daylight) and compared the results to the reference values provided by Babelcolor. Color patch C6 is not within sRGB color gamut and will be ignored.
Results


As I said my primary interest is in hues, if hues are wrong colors are wrong. I’m not the person to decide if it’s better that one color is vastly off when all the other colors are in line or when every color is slightly off. I´m sensitive for hue errors, other might be sensitive for saturation errors even if they are more easy to correct. It´s hard to pick a winner and this is no contest after all. I´m relying here on calculated CIEDE2000 Delta E errors. It´s a complex color difference algorithm and I won’t go any deeper into it, but it´s good to know that errors <1 are not perceptible by human eye and 100 when colors are exact opposite.



Sony A7 JPEG







Average Delta Error = 3,9
Worst color: A1 dark skin DE= 6,7


Authentic colors 


Sony A7 Adobe Standard Profile







Average Delta Error = 2,5

Worst color: B4 purple DE= 6,9


Saturation closest to reference values.

Sony A7 Profile made with Adobe DNG Editor






Average Delta Error = 3,0

Worst color: C3 Red DE= 3,9

Unsaturated yellow.


Sony A7 Profile made with QPCard






Average Delta Error = 3,4

Worst color: B4 purple DE= 8,8


Sony A7 Profile made with XRite






Average Delta Error = 4,4

Worst color: B4 purple DE= 12,0


Canon EOS 5D Mark II JPEG






Average Delta Error = 7,2

Worst color: A1 dark skin DE= 10,9

Compared to Sony more saturated and less bright colors. 

Canon EOS 5D Mark II Adobe Standard Profile







Average Delta Error = 3,3

Worst color: B4 purple DE= 7,1


Canon EOS 5D Mark II Profile made with Adobe DNG Editor








Average Delta Error = 2,9
Worst color: C3 Red DE= 7,1

Smallest hue deviation of the whole group, saturation also among the best.


Canon EOS 5D Mark II Profile made with QPCard






Average Delta Error = 3,7

Worst color: B4 purple DE= 9,4




Canon EOS 5D Mark II Profile made with XRite






Average Delta Error = 4,7

Worst color: B4 purple DE= 12,2