JPEG vs (8-bit) TIFF in scanning?

[ChrisR]

Hi again, I've just noticed that in my first experiments with colour negative scanning, I saved some images as both JPEG and TIFF. I use a Plustek 7500i with Silverfast SE. This is ancient Kodacolor II or X film (1971), for which there is no preset, so I was left experimenting with other presets. Kodak Gold is too blue, and I eventually settled on one of the Ferrania film types. Anyway, I wanted to give myself more leeway to edit later, so I saved as TIFF as well as JPEG. The resulting image sizes are 21 MB and 3.5 MB (I tend to scan at 2400 dpi, which equates to around 6 MP). What I didn't realise is that the settings I was using force 8-bit anyway, so this is a 8-bit TIFF.

My question is whether there is any advantage in keeping these TIFFs, since both they and the JPEG are 8-bit depth? There's no visible colour difference on my (uncalibrated) screen. The histogram is a bit narrower in the main peaks for the TIFF, and there's maybe a tiny highlight peak...

These TIFFs are not the 16-bit unprocessed ("raw") kind that you can feed into ColorPerfect, not that I have that sadly.

 

[joxby]

jpegs are compressed, Tiffs are a lossless file format......Viva la Tiff !

 

[ChrisR]

jpegs are compressed, Tiffs are a lossless file format......Viva la Tiff !

I understand about the difference in compression, but not about the implications of that for post-processing, particularly given the same bit depth. You've effectively already had the file compressed by losing 4 to 6 bits of colour depth of the image. Does the JPEG compression (which IS lossy but concentrates those losses in "less significant" areas, presumably like lower tonal variation) lose much more?

 

[Tricky69]

I suppose it depends on what you want to do with them 'whenever', and how much disk space you have to spare.

I had a 400mb Tiff printed at A2 for an exhibition and the detail really showed. The Jpeg was about 15Mb IIRC

http://www.flickr.com/photos/handmade-uk/5681711500/in/photostream/

On screen, you can't really tell the difference between the two.

 

[boliston]

I would always output to a TIFF initially as jpeg is mainly intended as a "final" file format rather than a format for further editing. I'd also go with 16 bits per channel rather than 8 bits per channel as I'm pretty sure the 7500i supports this.

 

[ChrisR]

I would always output to a TIFF initially as jpeg is mainly intended as a "final" file format rather than a format for further editing. I'd also go with 16 bits per channel rather than 8 bits per channel as I'm pretty sure the 7500i supports this.

Adrian, the 7500i does do 16-bit scanning. However, SilverFast SE gives me a stark choice here. If I want to save the scan at 16-bit I only get the linear scan, with no correction applied at all. If I want any corrections, Negafix, curves and levels etc, then SliverFast will ONLY save in 8 bits.

I have explored this as best I can, given my limited post-processing skill. Someone on TP helped me by taking a linear scan from an unsupported Fuji negative film (I think C200) and ran it through ColorPerfect for me; the results were great. But since I don't have Photoshop I can't use ColorPerfect myself.

I tried with some transparency film, comparing the results from visual inspection, JPEGs from SilverFast, and a linear scan 16-bit TIFF. The latter is strongly divergent in colour and needs a large amount of work (and skills I don't have) in PP to get it back towards what the actual slide looks like.

Here's an example of a picture taken in the Highlands. First the JPEG from SilverFast's conversion, with only minor level asdjustment in Aperture afterwards:

Here's the same image done with a linear scan and saved as a 16-bit TIFF (to be clear, I think the scan is the same in both cases; it's the post-processing in SilverFast that changes, and in this case there is none, not even gamma). Just to make it more obvious how different this version is, I've undone all of my adjustment attempts in Aperture in this image. The histogram is totally bunched up on the left hand side:

The slide looks very like the first image, and not at all like the second. With lots of PP work I can get part way there, but I've not been able to get results that are close enough.

Why was I bothering? I noticed that I wasn't getting shadow detail from some of these images, and I thought that might be a casualty of the compression in JPEG etc, so I wanted to use the 16-bit TIFFs to try to drag a bit more out. In the end I found that, with my set of skills, it doesn't work!

 

[dalex_257]

I understand about the difference in compression, but not about the implications of that for post-processing, particularly given the same bit depth. You've effectively already had the file compressed by losing 4 to 6 bits of colour depth of the image. Does the JPEG compression (which IS lossy but concentrates those losses in "less significant" areas, presumably like lower tonal variation) lose much more?

If you planning to do PP then loosing tonal gradations is not the thing I would want. It is much easier to cause posterisation in PP. You should look at it from this point perhaps - to get the bets out of PP it is better to start with a best possible data and JPEG is just not that. Also if your scanner supports 16bit TIFFs I would definitely save my scans as that for the very same reason.

 

[ChrisR]

If you planning to do PP then loosing tonal gradations is not the thing I would want. It is much easier to cause posterisation in PP. You should look at it from this point perhaps - to get the bets out of PP it is better to start with a best possible data and JPEG is just not that. Also if your scanner supports 16bit TIFFs I would definitely save my scans as that for the very same reason.

See post #6 in this thread for comment on the 16-bit issue. I was going to put that TIFF on my public Dropbox to see if anyone can show me ways to bring it back to closer to reality. However, when I looked more closely, I realised I'd saved it as a JPEG2000 with lossless compression, which seemed a good idea at the time (dropped the file size by 50%). What I hadn't spotted is that in the process SilverFast also dropped it back to 8-bit colour depth.:bonk: So really, it's back to square one...

If anyone is willing to have a go, I'll repeat post 6 with another image from the same set (for visual reference of the JPEG from SilverFast and the TIFF straight off the scanner) and give the URL to the 16-bit TIFF on my public Dropbox (41 MB)...

 

[dalex_257]

If anyone is willing to have a go, I'll repeat post 6 with another image from the same set (for visual reference of the JPEG from SilverFast and the TIFF straight off the scanner) and give the URL to the 16-bit TIFF on my public Dropbox (41 MB)...

I am using the VueScan or Epson software (for my scanner) and neither definitely does JPEGs only. I have never used SilverFast but I am somewhat surprised that it cannot produce non-linear 16bit TIFF version (unless I am misunderstanding what you are saying). My understanding before was that SilverFast is roughly equivalent to VueScan in functionality hence the reason it is a choice of so many pros.

Workflow wise, I am not really into ColourPerfect and linear scans so I try to get it right from the scanner mostly (with limit of course - I try not to adjust brightness or curves there and do it in PP). For transparencies if you want to get the colours just as they look on film I would advise to get something like a Wolf Faust film targets and profile your scanner to eliminate all the casts introduced by scanner itself. I got myself S1 pack for all the films and profiled Velvia for and Kodak slide films and all the cast is gone. All I do now is adjust exposure, tweak the curves in PP to boost contrast and sharpen.

 

[ChrisR]

I am using the VueScan or Epson software (for my scanner) and neither definitely does JPEGs only. I have never used SilverFast but I am somewhat surprised that it cannot produce non-linear 16bit TIFF version (unless I am misunderstanding what you are saying). My understanding before was that SilverFast is roughly equivalent to VueScan in functionality hence the reason it is a choice of so many pros.

Workflow wise, I am not really into ColourPerfect and linear scans so I try to get it right from the scanner mostly (with limit of course - I try not to adjust brightness or curves there and do it in PP). For transparencies if you want to get the colours just as they look on film I would advise to get something like a Wolf Faust film targets and profile your scanner to eliminate all the casts introduced by scanner itself. I got myself S1 pack for all the films and profiled Velvia for and Kodak slide films and all the cast is gone. All I do now is adjust exposure, tweak the curves in PP to boost contrast and sharpen.

Thanks Alexy. There are several levels of SilverFast; I have SE Plus (came with the scanner) which is mostly OK for my needs. To do scanner profiling and other fancy things (probably including 16 bit nonlinear TIFFS!) I'd have to spend another pile of cash to upgrade to SilverFast AI. I've pretty much decided it's not worth it for the job in hand, especially since I can get better scans with processing for my current films. I'm keeping the 7500i and SilverFast SE for the backlog of films from the "children years", for which great quality is usually not so important (though my wife may disagree...).

When I first started having colour cast problems with early colour negative film like Kodacolor-X I did get a trial of Vuescan, but TBH it didn't seem to be doing the job any better. Maybe if I'd stuck with it and learned the (very different) interface better, I'd have got better qulity results!

 

[dalex_257]

When I first started having colour cast problems with early colour negative film like Kodacolor-X I did get a trial of Vuescan, but TBH it didn't seem to be doing the job any better. Maybe if I'd stuck with it and learned the (very different) interface better, I'd have got better qulity results!

Color casts in negative is something that is rarely addressed by scanner software - you need to do it in PP with curves. The reason I personally picked VueScan is that it allows you to vary your starting point for PP and switch off all the scanner software "optimisations" (so you get the sharpest image). And of course the fact that it does not cost an arm and a leg . You can get linear raw scans (for ColorPerfect) or reversed negative with initial WB tweaks to get the colours close to what they should be and finish them up in PP. For the postprocessing of the negatives and colour correcting them there is a multitude of the approaches - this one for example (requires paid subscription to see the whole article) which is by far the easiest approach. I usually use the curves and colour correction by Dan Margulis.