Camera-Scan Challenge for Color-Neg, as Automatic as Possible

[charjohncarter]

Assuming, you can achieve a acceptable color negative inversion, I am wondering where pixel shift fits into this discussion: re: 'saturated grain particles and digital noise.' I have a 36MP FF DSLR cum pixel shift that I can use on my copying device.

 

[jzagaja]

And how it compares to 1pixel scan from drum scanner. I've sent samples to Cinelab in US.
They use true 16bit HDR - best lens and sensor. Since we have ready kinematics for making xy scanner it would be interesting build scanner based on cheap camera, microscope lens with computational photography (astrophotography algorithms) or hiend low res sensor from eg Hammatsu.

 

[charjohncarter]

And how it compares to 1pixel scan from drum scanner. I've sent samples to Cinelab in US.

I would very much enjoy a reasonable 'study' on this: not a bunch of home grown made suppositions without proper statistical review. And in the end come to a conclusion that is, as a OP that started this thread (maybe I'm guessing), what is a reasonable .

 

[jzagaja]

My experience is: Fuji X-trans sensor not good for 5000dpi scan of 35mm. A6300 with good macro lens can 5000dpi easy and lot of high frequency data.

Worn drum scanner (register error) - worse than above:
https://flic.kr/p/2etjoJ2


Good working drum scanner, great control of grain aliasing, less halos:
https://flic.kr/p/2bt3fVB

but most are worn today and problem with dynamic range - 12bit ADC at best.

https://flic.kr/p/2a6btzo

 

[shijan]

Fuji X-trans sensor just need quality debayer method. A lot of raw convertors use very basic debayer for X-trans sensors and image looks really ugly at pixel level. It also depends of post sharpening. Note that i done all my tests with no any sharpening at all.

Anyway i could be very interesting to see side by side tests of X-Tras vs Bayer tests at the same resolution and same film scans.

 

[jzagaja]

Here I've shown in-camera processing and slide digitization.

 

[jzagaja]

Make Tiff + DNG 12.3 makes fairly good job. I just applied LOG curve, removed orange mask, inverted and applied Photoshop 7 function auto color. Saturation natural (on wide gamut OLED screen) without clipping, contrast neutral.

 

[charjohncarter]

Looks good, and it goes well with the OP's request of as automatic as possible.

 

[jzagaja]

And test box same way, orange mask removal not needed. I'm using channel mixer to change saturation vector close clipping and prevent banding.

 

[Michalm]

How is everybody feeling about Negative Lab Pro and Grain2Pixel software ? To me a lot of the time results look like what comes out of VSCO lightroom presets , still Grain2pixel looks results look more natural .

 

[jzagaja]

Two Portra drum scans with this method. (watch on HDR display)

https://www.flickr.com/photos/66109016@N04/51116127966/sizes/m/

https://www.flickr.com/photos/66109016@N04/51116038158/sizes/m/

 

[shijan]

Make Tiff + DNG 12.3 makes fairly good job. I just applied LOG curve, removed orange mask, inverted and applied Photoshop 7 function auto color. Saturation natural (on wide gamut OLED screen) without clipping, contrast neutral.

Nope, it is not look good. You lost all saturated colors here, and overall photo look dull. Red backpack and especially red stripes should be extremely saturated, but not clipped. When i look on film negative in real life those colors are incredible high saturated acid green compare to all other colors on the frame. This is why i choose this frame for tests. Film scans with low saturated colors are way easier to process, because they more-less fit to sRGB.

Also beware that MakeTiff don't do any color management. It just output debayered tiff without any proper camera input profile. Option to attach some basic ICC profile from the list in MakeTiff makes things even more confused and incorrect because instead of proper input camera ICC profile it attaches incorrect basic profile :bang:.
Develper of ColorPerfect "explained" to me that his "sysmem" designed like this because ColorPerfect attempt to guess input color based on some film preset and autolevels. And in the end he suggests to do all transformations in sRGB because it is close to input color space of old scanners 🙂
This is totally stupid and technically incorrect way to deal with colors. So good luck to everyone who still use MakeTiff...

 

[jzagaja]

Saturation vector is camera filters + RAW converter related and has nothing to do with this method. For example Red cameras and Scanmate drum scanners deliver near clipping colors. This is arbitrary - you can change saturation vector with 3x3 matrix via ICC over reflective target photographed on film.

Your conversion shows typical effect of processing colors in perceptual space (non-linear), over bright colors - even Fujifilm doing this in camera processing.

Below reflective target and drum scanner with above method:

 

[jzagaja]

And example of straight debayered image from Blackmagic camera and after applying 3x3 saturation vector via ICC profile from reflective target.

 

[shijan]

And now compare overloaded colors in your chart to colors that you can get with proper custom-made ICC profile:

 

[jzagaja]

Like I wrote you can adjust saturation vector computing ICC profile. Drum scan is without any profiling. We don't know what Color Quartet soft is doing.

Profiling scanner with transmittance target to scan negatives is questionable especially 3D LUT profiles that induce banding. Try photograph transmittance or reflectance target on film.

 

[ColSebastianMoran]

I'm the OP... I'll ask that we not argue about things. Jack and Shijan, you both obviously know a lot and have access to very good resources, you both have methods that work for you, and we all appreciate you describing your methods and showing results.

Aim of this long discussion thread is to provide info that will help others with camera-scanning of color negative materials.

 

[charjohncarter]

I'm the OP... I'll ask that we not argue about things. Jack and Shijan, you both obviously know a lot and have access to very good resources, you both have methods that work for you, and we all appreciate you describing your methods and showing results.

Aim of this long discussion thread is to provide info that will help others with camera-scanning of color negative materials.

Thanks, Sebastian, I still would like examples of workflow from digital raw files to a finished color positive file that is not only repeatable but has some adjustability.

 

[ColSebastianMoran]

Shijan and Jack: Looking over your posts, I'm not clear on your process. Could you list your steps. For example, here's my list for an earlier not-too-bad process in Photoshop:

- Shoot RAW
- Levels adjustment to normalize the histogram for each channel (pretty wide, no clipping)
- Curves layer to invert, put a big bend in the curve
- Curves layer, Option-Auto, snap neutral midtones (gives an auto WB)
- Adjust the first Levels layer for minor adjustments, shift colors as needed.

Several other processes are mentioned in the previous pages of this thread.

Would like to understand both of yours.

 

[shijan]

I guess i already posted my step by step workflow in details and also record two short videos. It may look slightly random, but i guess soon or late i may compile it to more compact size and put it to separate thread.

Note that it is impossible to fit all workflow to some basic simple 1-2-3 steps, because there is a huge amount of nuances and dependencies that you need to understand.

For example "Shoot RAW" step that may look simple, in reality consists of huge amount of dependencies and nuances:
Exposure settings.
Proper histogram monitoring to avoid clipping
Light source quality
Calibrate WB to light source
Custom-made camera input ICC profile
App used for Debayer
Remove build in contrast curve form raw file.
Debayer quality
Problem with sharpness, antialiasing and noise reduction settings
Export to Tiff with Linear gammma
Color management system (AppleCMS vs AdobeCMS vs LittleCMS) affect final look.

and so on...