Hints for Linux workflow?

[barnacle]

Evening all,

I'm a long-term Linux-only user, with an ancient Epson 2450 Photo Perfection scanner and a cardboard(!) holder for 5x4 monochrome negatives.

It appears that the 2450 can read a 16-bit mono negative image with the Epson drivers - though I have not yet checked that it actually sees more than 256 gray-scale levels. At present I don't think most of my negatives are using the full contrast depth of the system, so more work required there.

Anyway, the plan is:


Scan the image
Save in a 16-bit lossless format - TIFF at a guess
Import into Darktable
Bend the curves as appropriate - in particular gain and contrast
Remove scan defects and dust
Save, with a converted 8-bit version for web viewing



Does this sound reasonable? Can anyone point me at an easy starter for Darktable?

(By way of background - I had thirty years in broadcast TV, which means that my understanding of gain, contrast, and gamma is perhaps the complete antithesis of everyone else's in the photographic world!)

Neil

[RSalles]

Neil,

The your workflow seems reasonable for me, I would export from Darktable to 16 bits tif continue from that converting it in the very first step to 32 bits for further editing.
Darktable is great for noise reduction and sharpening with great options as algorithms for both, but as soon as it's exported I find its result too "strong" compared too that visualised in the software itself. I use to be very conservative there, fine tuning with Gimp in 32bits and re-exporting.
Required: 2.9 Gimp version https://www.gimp.org/news/2015/11/27/gimp-2-9-2-released/

Cheers,

Renato

[Harold_4074]

I have an Epson V750 at work, and use the Epson software to import into Photoshop. I always used 16-bit grayscale mode unless my (old) Photoshop would perform some operation only in 8-bit grayscale. Recently, I found myself writing code to do image manipulation, and noticed that the 16-bit files didn't take noticeably longer to process than the 8-bit ones. Checking the actual file size, I was surprised to discover that the difference was only a factor of two for uncompressed TIFF. Apparently, the older Epson software allows a choice between 8-bit data and 16-bit format data, which Photoshop notices but which is otherwise pretty much irrelevant.

This is probably old news to everyone else, but it made me blink :)

[barnacle]

Thanks guys.

This scanner is doing my head in; I can't make it do the same thing twice. Basically, it seems to be scanning a negative and then ignoring all the XSANE controls when it inverts it; it's very fond of extremely pale negatives, but it's not consistent. Scan the same negative today and tomorrow and there will be different results.

I'm an old broadcast engineer. What I want to be able to do with the scanner is to be able to select peak white and peak dark on the negative and tell it to grab the result between those points, linearly, and then invert the result. Apply gamma after that in whatever software is handy. I'd like a proper waveform monitor, such as is used in broadcast but which I have never seen in any image manipulation software outside a studio, so I can see what bit of my image is doing what. A histogram is less than useful; I care about the picture, not the statistics.

/rant

145290

Harold, why would you expect a 16 bit image, uncompressed, to be more than twice the size if an 8-bit image? I would expect the processing on a modern processor to be be either the same time (ignoring loading time) or faster for a deeper bit size. I can't see anyone not promoting the sample values to anything other than the current 32 or 64 bit integer size, if only to simplify processing. (Er, except it might be even better to promote to a 64-bit float, as Darktable does).

Neil

[Neal Chaves]

I never heard of Darktable, so I just downloaded it to try in my Red Hat Fedora 23 system. I scan with SANE on old UMax SCSI scanners and then adjust the files in GIMP. I have my color prints made on a Fuji Frontier machine at a local department store known for low prices. They only print from 8 bit jpegs so I don't bother with the 14 bit I could obtain. The prints look good, I think. A 6X9 cm color neg or transparency at 3000 dpi yields a 96Mb jpeg and the Fuji machine has no problem with that and make a nice 12X18" in an hour for $6.

[RSalles]

Neil,

Suggestion: export the file scanned after the adjustments you can do with consistency to TIF 16bits. Instead of Darktable - as it's intended to work mainly with raw files not bitmap as tif - do inversion and curves and the rest needed in 16bits tif with the software you usually do (Photoshop,Gimp or other),

Cheers,

Renato

[Harold_4074]

why would you expect a 16 bit image, uncompressed, to be more than twice the size if an 8-bit image?

I wouldn't--I mis-stated the phenomenon: when I unpacked the data using Mathematica, the histograms showed only two more bits of depth, not eight more. The file size/processing time issue had to do with color images, which of course should be exactly (to within some header information) three times the size of an 8-bit file. The combination of older Photoshop (CS4) and even older Epson Scan (3.81) has exhibited quirky behavior in other ways, many of them not readily reproducible. I should try the .tiff export again and see what happens.

Sorry for the confusion---I must be older than I thought :)

[barnacle]

Cheers Renato,

I'm exporting to TIFF 16 bit, but I'm still not convinced it's actually using the full range, or anything like it. I need to take one of the image files to bits and have a look inside!

Darktable is a recent experiment, and I'm not sure whether it's the thing for me, but my Gimp is 2.8.10 which down-converts a 16-bit TIFF to 8-bit. 16-bit is announced for 2.10...

But it's all a bit moot unless I can get the scanner to behave.

Neil

[barnacle]

@Harold - by bad - I failed to read your comment properly. Sorry about that.

Neil

[Harold_4074]

Neil--

No, the error was on my part. In my mind, I had crossed up the processing time difference for color-mode black-and-white images (these are scans of x-ray film) and monochrome. The bit-depth issue is separate; a clue may be present in the fact that Epson Scan allows a choice of 8- or 16-bit images, but the histogram adjustment function (which I normally do use) only reports 256 levels. I should try a scan in 16-bit mode with only default levels and see if that makes any difference.

The processing that I was doing is probably a warped version of what you want---I needed to take films developed with no particular process control and make images which would allow me to locate features containing only a narrow range of values. But--and this is the tricky part-- without any subjective input through Photoshop. Using Mathematica for this is like using a steam shovel to bury a dead rabbit, but it's what I have at hand :)

[RSalles]

Cheers Renato,

I'm exporting to TIFF 16 bit, but I'm still not convinced it's actually using the full range, or anything like it. I need to take one of the image files to bits and have a look inside!

Darktable is a recent experiment, and I'm not sure whether it's the thing for me, but my Gimp is 2.8.10 which down-converts a 16-bit TIFF to 8-bit. 16-bit is announced for 2.10...

But it's all a bit moot unless I can get the scanner to behave.

Neil

Neil,

Gimp 2.9 is capable of 32 bits as it uses the GEGL graphics library for image processing. My bet is you could easily install and use the beta 2.9 - I did for several months ago and haven't had a single crash,

BTW, the developers has pointed that using 16 bits image instead of 32 gets the same CPU and mem load with the new Graphic library, I made myself the load test and could observe that too,


Cheers,

Renato

[Ken Lee]

Is the 32-bit version (2.9) currently only available for Linux ?

(I can run Linux on a VM under OS X. What Linux distro would you recommend ?)

[RSalles]

Ken,

I don't have a Windows PC running here but my bet is it's also available for Windows,

145330

Link for download:

http://nightly.darkrefraction.com/gimp/

If you're searching for a stable Linux distro with a nice user interface, try out Mint Linux (I'm running the latest 17.3 version). Very stable, based upon Ubuntu Linux,

Enjoy,

Renato

[barnacle]

Mint 100% - it's by far the simplest to use for anyone who's come from the Windows world (as in, it still has a hierarchical menu system and a system tray). This is a Mint 17.3 installation that replaced the Chromeos on a Toshiba CB2 - not an easy job but I've documented it on the mint forums. Lovely screen, but a bit short of memory. Better (for me) as a full mint machine than as a chromebook!

I'm slowly getting somewhere with the scanner. It seems the results are best as a full-colour image, *not* as a negative. But it's still quantising in the thinner parts of the negative, in spite of 16 bit TIFF output. I'll look into Gimp 2.9.

Neil