[normhead]

But I think those numbers refer to two different things - the 12/14 bit analog to digital converters (one per pixel, contained within the sensor) in the cameras are preprocessing the data before any file structure is in the works

But what if you are wrong.
The issue in jpeg vs raw is the number of gradations in each colour range the can be displayed. Jpeg has on only 8 bits in the black spectrum. You can display 8 shades of black.

There are way more gradations in nature than you can capture with any camera. Imagine how detailed an image would be if you could look at a 20,000:1 scene and look at every photon. You can't and it's not necessary. You only need to capture what you can see. The biggest difference between raw and jpeg is in the deep shadows end. In the high end 8 bit is probably sufficient.

My highest quality jpegs look very close to my raw files. The jpeg tries to save everything important in and for the most part is successful. The issue is, what happens if you want to boost the shadows. You have 2000 gradations in each 14 bit channel as opposed to 16 in the jpeg which allows you to make much more realistic looking shadows.

Once you are done, and have it the way you want, save it as a jpeg, but use a non-destructive program so if you decide you want another edit, you can work from the most available information possible. Once you've committed to jpeg, your hands are tied, you get what you get.

An off camera jpeg can look just as good as the raw, until you decide you want more shadow detail. As pointed out above, even in jpegs, there's lots of room to work in the high end. It's the shadows where raw proves it's worth.

I'll put some examples together for you if I get a chance, but it may be a while.

---

Last edited by normhead; 11-18-2018 at 04:18 PM.

Here's an interesting article on the subject:

JPG vs. RAW, 8-bits vs. 16-bits, sRGB vs. wide-gamut RGB. Which to choose, when and why

And another:

https://www.kenrockwell.com/tech/raw.htm

[slartibartfast01]

And the "jpeg files contain 8 bits" is incomplete - JPEG is twenty-four bits, eight bits for each of the three primary colors. They also store three or seven bits per pixel of luminance information (one nybble or byte, respectively, containing a signed binary integer), as I understand it, for a total of twenty-four or thirty-two bits per pixel.

I thought that the 8 bits per colour channel in a jpeg file contained all the information required to specify a pixel colour and brightness. So there is no need for an extra three to seven bits per pixel of luminance information. Luminance can be calculated from the RGB values.

[microlight]

JPG files are 24-bit in the same way that RAW files from my K-3II are 42-bit! What we’re also missing is the point about compression, which isn’t actually the issue. Even DNG files are compressed to an extent, but here what we’re talking about is lossy compression (as opposed to lossless compression) which is what happens in camera JPG engines. There is an option in the JPEG protocol for lossless but it’s rarely implemented. So JPG files from our cameras are lossy (data are lost) whereas RAW files are lossless.

[BrianR]

I have a couple of observations about the JPEG's being produced - first, JPEG isn't an eight-bit standard; it's a twenty-four bit standard, eight bits per color channel, red, green, and blue. Secondly, it isn't necessarily a compressed image - there's no loss of data if you select the largest file size of JPEG the camera can store, because that's the "100%" standard (uncompressed). The K-1 records the same amount of data for both the TIFF version (whether stored as DNG or PEF), as it does for the uncompressed JPEG: L(36M:7360x4912); RAW: (36M:7360x4912) (Specifications | PENTAX K-1 | RICOH IMAGING). I have no reason to believe that it isn't the SAME data..

Same viewing resolution does not mean the same data. The JPEG compression (even at the max star rating of the camera or whatever photo application you're using) is a lossy compression, so it's not the same data. Yes, there is a lossless version of jpeg, but afaik, nothing out there uses it. Take a look at the steps involved in encoding a jpeg at JPEG - Wikipedia to see a rundown of the steps, the downsampling and the quantization step involve throwing information out. The highest jpeg star level on many pentax cameras use the quantization matrix corresponding to quality level of "100" in the independent jpeg group standard (which is still lossy) and 4:2:2 downsampling.

You lose data by going to a jpeg, there's no going back. The data thrown away in the jpeg conversion process is done intelligently though, and at low compression levels for most photographic images the visual impact is insignificant at the highest quality setting. But that doesn't mean you won't get into trouble when it comes to heavy editing.

[stevebrot]

Digital sensors are actually analogue devices, these voltages are quantized through the ADC. And the qualities** of the ADC's have a tremendous impact on what is recoverable - I just thought I'd mention that for clarity.

So true! My posts are too wordy as it is, so I figured that it is best to skip some of the thornier details and consider the sensor to be the collector + ADC. After all, the data we work with are digital as are the practical constraints on the way to making a viewable image, so that makes a reasonable starting place.


Steve

[TonyW]

I thought that the 8 bits per colour channel in a jpeg file contained all the information required to specify a pixel colour and brightness. So there is no need for an extra three to seven bits per pixel of luminance information. Luminance can be calculated from the RGB values.

I believe that this is correct. In Photoshop terms we think of 8 bit, 16 bit or 32 bit images. The common JPEG can be thought of as 8 bit (or 24 bit;8+8+8 bit). JPEG 2000 can use higher bit depth but is a format that does not appear to have taken off in any way.

So a JPEG can address up to 256 levels of luminance per channel to represent any one of 16,777,216 colours per pixel.

A 14 bit capture can address up to 16,384 levels per channel to represent up to 4,398,046,511,104 (4.39 Trillion)

Pretty staggering numbers and the reason you need (or may need) is that by working in higher bit depth it is possible to eliminate colour degradation that may occur in image manipulation such as banding. Sometimes there is just not enough headroom in a JPEG image for manipulation and the image can break, whereas a 12 or 14 bit raw will not

---

Last edited by TonyW; 11-18-2018 at 12:28 PM. Reason: cannot do maths correctly :-)

Here's another definitive (for my purposes) reason for a selection:

https://www.pentaxforums.com/forums/58-troubleshooting-beginner-help/53546-w...tml#post517719

[TonyW]

Here's another definitive (for my purposes) reason for a selection:

StevenVH - View Profile - PentaxForums.com

Sorry unable to make any sense of this including the link to someone’s profile ? Edit Spoke too soon

---------- Post added 11-18-18 at 11:15 AM ----------

Still sorry, maybe I have just missed the obvious but it is not clear what point you are trying to make. Could you try and clarify please

[TonyW]

Ok, so I found the article that I read a few years back. It's in Russian, so I'll try to quickly break down what the guy is saying.

Basically, he's comparing the LR/ACR raw conversion to an app called RPP (Raw Photo Processor), which is an open source converter, I'm pretty sure it's dead by now in terms of development.
He's got image samples that show comparisons between the two, the RPP seems to create a more natural red color that isn't blown out.

?????? ?? ????? ?????? ??????????? ????????? RPP ? ??????? ????????????: dmitry_novak

He mentions that the white balance is correct (I'm guessing he metered it with a grey card, I believe he's mainly a still life/product photographer).
I'm not sure how this applies to the latest LR/ACR, I don't know if the basic color profiles were further calibrated or not.

Here's a shot I just played around with in Capture one, I could not make it work at all in Lightroom, at all. It's not perfect right now either, but I think with some further work it'll be ok. I actually might switch to Capture One lol, the only thing holding me now is the Photography package value from Adobe (I use Photoshop a lot), but I might just bite the bullet and get the capture one on top of it. The lightroom sample is the weird looking one (again, I couldn't make it work. If you want, I can upload the raw for you to test).

Thanks for going to the trouble of finding the article. I am familiar with the work of Dimitri Novak and Raw Photo Processor and trust his judgement on these things. So if he said x camera not good in Adobe I would suspect him to be correct. Never tried the software as I believe Mac OS only. Last version seems to be over 4 years ago so not sure if it is still being supported.

Looking at the Adobe JPEG vs the Capture One and having a play with just this limited data version I suspect that ACR can approach your preferred rendering with a minimum of tinkering. COne has a reputation for warmer (orangey skin tones) as default that some prefer. I do not like COne for the selfish reason that they are blocking my medium format files from 645 as they compete too well against Phase One etc.

Would be happy to have a play and see how close I could get to your desired rendering - no guarantees of course COne may just prove better for this image but it would surprise me. Just pm me with Dropbox or similar

[Digitalis]

a JPEG can address up to 256 levels of luminance per channel to represent any one of 16,777,216 colours per pixel.

Just don't forget that with a Bayer equipped RGBG sensor the Red and Blue channels are interpolated with 50% of the cameras linear resolution being in the green channel.

There have been experiments with different bayer patterns - Sony and Fuji have produced sensors with unusual CFA's before - The biggest problem is controlling metamerism and channel cross talk. There have been attempts to include an unfiltered photosite RGBL - the issue with this is having a dedicated unfiltered luminance channel was prone to clip really easily relative to the colour channels, so the sensor designer made the luminance pixels really small - which caused problems with diffraction. The luminance pixel also responded to different spectrums of light in aberrant ways which threw off the Delta-E of the sensor entirely...

[TonyW]

Just don't forget that with a Bayer equipped RGBG sensor the Red and Blue channels are interpolated with 50% of the cameras linear resolution being in the green channel.

There have been experiments with different bayer patterns - Sony and Fuji have produced sensors with unusual CFA's before - The biggest problem is controlling metamerism and channel cross talk. There have been attempts to include an unfiltered photosite RGBL - the issue with this is having a dedicated unfiltered luminance channel was prone to clip really easily relative to the colour channels, so the sensor designer made the luminance pixels really small - which caused problems with diffraction. The luminance pixel also responded to different spectrums of light in aberrant ways which threw off the Delta-E of the sensor entirely...

Good points, and I know some find it hard to accept that the nature of the beast is that it cannot deliver a full resolution of the sensor so from our 36 mp sensors we can only expect up to 50% real resolving power.

I know that Fuji pioneered some adventurous arrays. I still have. Fuji E550 that IIRC had a tilted hexagonal arrangement? Again if memory serves they produced the S5 SR Pro with the Super CCD Pro array @ 12 mp. According to many wedding photographers a fantastic step forward in IQ able to hold more detail in brides white dresses while still holding great detail in deep shadow areas. Once some photographers realised that Fuji used 6 million of the S type photodiodes to capture main image information and 6 million of the smaller R photodiodes for bright area information there was quite a bit of net chatter. Claims of being cheated “Camera is not 12mp it is only 6mp”. A view I could not share

[awscreo]

Thanks for going to the trouble of finding the article. I am familiar with the work of Dimitri Novak and Raw Photo Processor and trust his judgement on these things. So if he said x camera not good in Adobe I would suspect him to be correct. Never tried the software as I believe Mac OS only. Last version seems to be over 4 years ago so not sure if it is still being supported.

Looking at the Adobe JPEG vs the Capture One and having a play with just this limited data version I suspect that ACR can approach your preferred rendering with a minimum of tinkering. COne has a reputation for warmer (orangey skin tones) as default that some prefer. I do not like COne for the selfish reason that they are blocking my medium format files from 645 as they compete too well against Phase One etc.

Would be happy to have a play and see how close I could get to your desired rendering - no guarantees of course COne may just prove better for this image but it would surprise me. Just pm me with Dropbox or similar

Sure, I'll upload the dng for you tonight. I noticed a massive difference in results I'm getting from capture one, I opened some of the images that didn't look great in lightroom and found it quite easy to tweak them to be pleasing enough in capture one. I think I'm going to switch my library to capture one over weekend. Also really really need to clean up my files lol, I have too much junk trash that I will never touch anyway

[PDL]

And another:

https://www.kenrockwell.com/tech/raw.htm

Ken Rockwell? Really? The guy who pushes saturation to 11 and who pushes post processing software (Perfectly Clear for example) out the yin-yang?

Even Tony Northrup would be someone better to quote or:

Why shoot RAW? | Capture One Blog
5 Reasons to Shoot in Raw Instead of JPEG
Why Shoot RAW - The Top Reason to Shoow RAW!
etc. etc. etc.

Also remember that LR, Darktable, Capture One and other RAW processors do not use the same demosaic algorithms. There are differences in ACR version to version too. When you use OOC JPEG's you get the demosaic process that the camera engineers produced - and they do not change for that camera body ever. When using RAW converters it is almost the same as using different film emulsions.

For example you could say that ACR => Fujichrome, Capture One => Kodak chromes (but not Kodachrome ) and as for others most likely Agfa and some off shoot brand stuff. You might want to give this site a look see too: Cambridge in Colour - Photography Tutorials & Learning Community far better than anything on KR's site.

When you shoot RAW, you are in control of the processing, not some engineering team from (fill in how many years ago your camera was "designed"). Learn to be the decision maker.

[sbh]

Be sure to read the initial post carefully. I quote;

"It is important to note that all Jpg processing was off during the shot (DNG) and that no Jpg processing was turned on during the conversion in the camera of RAW > Jpg. (Yep... just to be clear the shot itself was just a single RAW DNG file, not a RAW+ where a Jpg is created at the time along with)."

The Jpg 'edit' the camera (K1) made is as close to the RAW file itself, no Jpg processing on at all. My own 'edit' was not biased to one image or the other, just extreme Highlight and Shadow recovery made to both files, with some slight exposure boost;

Exposure +0.83
Contrast +43
Highlights -100
Shadows +100
Whites +29
Blacks 0

My issue is not about Jpg with tweaks in the camera vs RAW (which is often the battle), my issue is a DULL non processed Jpg from camera that is as close to being a 'RAW jpg' as you can get, would appear to have greater room to breathe in terms of recovering exposure, highlights, shadows and details. It doesn't make a lot of sense.



...

...

I don't use ACR, I do use LR>PS and have seen the PS Saturation boost it can often apply, so I do hear what you're saying.

However in this particular example, reducing Green Saturation by -20 still leaves brighter 'more prone' to highlight clipping than the Jpg. I don't think this issue has anything to do with colour levels.

...

So, if I'm hearing everyone correctly then what's going on is one of two things (or a combination of the two);

1) The RAW and Jpg are not even starting off on even footing, LR is interpreting the two files differently (upon import) and as such making global adjustments to both is unfair (depending upon the adjustments and which direction the adjustments are favouring the other). I should try RawTherapee (it's the only other RAW editor I have installed).

2) The RAW file may be 'encountering problems sooner' than the Jpg version as the extra data things it has over the Jpg are the things that are triggering the issues i.e. the additional bits of information that the RAW file has over the Jpg are the things we're seeing causing the problem. The Jpg can't trigger the same issue as it simply lacks the data in the file that could get 'triggered'. In some situations it may be beneficial to have those additional bits of information, just not in this case (if the sliders were actually the look the photographer was aiming for). This time the image is better suited to its lossy Jpg counterpart, where it's missing information is actually of help, not hindrance. Push the sliders differently and the opposite could be true.

Is this right?

No absolutely not. I tried to make this comparison as fair as possible, anything Jpg related was turned off for the shot as well as turned off for the RAW extract. Dull and lifeless Jpg as much as possible.

Basically what you are doning is comparing the results of two different raw converters. One is the in-camera software (i believe the same engine as DCU) and the other one is LR.

So to me it is indeed uneven grounds you ares starting from. The jpeg has the DCU default curve + the "natural" edits applied, the raw has the LR curve applied. Add another raw converter and you'll get yet another result. (There is also a "flat" profile in the K-1 which is even flatter than "natural".)

Afaik, when viewing in a raw converter, a jpeg doesn't have any adjustments added as it is already processed and therefor just dispayed as it is – the raw otoh has adjustments applied in order to display a starting point before the user makes any adjustments.

For a "headroom" comparison I would try to push shadows and highlights as far as possible and compare how far it goes on either file.