 |
 | Home | Contact Us | FAQ | Search & Site Map | Link to Us |
 | Sign In | Join | Other 45 Sites in Network |
| PhotoKB HomeDiscussion Groups Digital Photography Digital PhotoDSLR CamerasZLR CamerasPoint & Shoot CamerasFilm Photography 35 mmLarge FormatMedium formatDarkroomFilm and LabsOther EquipmentPhoto Technique Nature PhotographyPeople PhotographyTechnique GeneralGeneral Photo Topics General TopicsAustralian PhotographyUK PhotographyDirectoryPhoto Clubs |
Photo Forum / Digital Photography / DSLR Cameras / April 2006ACR 16 bit raw photo conversion
[x-posted to alt.graphics.photoshop & rec.photo.digital.slr-systems] I discovered (perhaps misguided?) that if I zero out all the adjustments in ACR (Adobe Camera Raw) except white balance, and save as 16 bit, the result has enough information to do anything that ACR will do for contrast, saturation, etc. I'm using CS (not CS2). For high contrast images, I used to do two conversions for highlights & shadows & mask those together but it can be done with one straight 16 bit conversion & masking adjustment layers with a lot more flexibility. I had thought that the ACR adjustments were working on some magical RAW level of high bit math but in fact, 16 bit (vs 8 bit) holds everything there is to get. I'm familiar with holding down the alt key in ACR while sliding exposure and shadows and that's a nifty tool but why didn't they just put it in the core PS adjustments and why these new terms like shadows & brightness? On a related note, ImagePlus http://www.mlunsold.com/ is a program used for astrophotography which features a higher bit math, which is reported to perform cleaner conversions. From their descriptions: "- Read and write 8, 16, and 32 bit integer or 32 bit float RAW images with one or 3 color channels. Interleaved or non-interleaved channels are also an option. - Read and write 8, 16, and 32 bit integer or 32 bit float and 64 bit double FITS image formats." The argument I heard is photoshop *only* does 15 bit math and this does 64 bit math so there is less rounding error and cleaner results, especially in the shadows where the amount of information is minimal. Then I hear that camera files are really only have 12 bits so 15 should be plenty, right? I'm not clear what 32 bit or 64 bit even means. One issue may be that RAW data is linear, not gamma adjusted, so when you do the gamma adjustment to 2.2 that stretches the data compressing highlights and expanding shadows but expanding won't damage anything and the highlights can usually afford to be compressed some. Strictly speaking this expands from 4096 levels to about 13000 levels or about 3x if you don't do any compression. I've also read that photoshop has limitations to what can be done with high bit images such as 16 bit & 32 bit like no adjustment layers. I think this is outdated and not an issue in CS. I don't see any option for 32 bit files though. Any insights on these isssues? > [x-posted to alt.graphics.photoshop & rec.photo.digital.slr-systems] > > I discovered (perhaps misguided?) that if I zero out all the adjustments > in ACR (Adobe Camera Raw) except white balance, and save as 16 bit, the > result has enough information to do anything that ACR will do for > contrast, saturation, etc. I'm using CS (not CS2). ... > Any insights on these isssues? IMHO 8 bits has plenty of headroom for editing, and I have yet to see any exceptions to this in images using "normal" colorspaces. That said, it's a fairly complex issue, and reasonable people differ. The question is becoming moot as processors get faster and memory gets cheaper. Here are some more insights that you might find interesting. http://mike.russell-home.net/tmp/DanMargulis16Bit/  SignatureMike Russell www.curvemeister.com 8< > Here are some more insights that you might find interesting. > > http://mike.russell-home.net/tmp/DanMargulis16Bit/ Wonderful read that, Mike. Many thanks! Wooohoooo. I feel L I B E R A T E D! Those 76Mb 16 bit ProPhotoRGB TIFFS are now history. :-) Nigel > 8< > [quoted text clipped - 4 lines] > Wonderful read that, Mike. Many thanks! Wooohoooo. I feel L I B E R A T E D! > Those 76Mb 16 bit ProPhotoRGB TIFFS are now history. :-) ProPhoto is *not* one of those 'normal' color spaces Mike is talking about. If you use ProPhoto in 8 bits, you will risk banding. SignatureJohan W. Elzenga johan<<at>>johanfoto.nl Editor / Photographer http://www.johanfoto.nl > ProPhoto is *not* one of those 'normal' color spaces Mike is talking > about. If you use ProPhoto in 8 bits, you will risk banding. Yes, I realise that. I was persuaded once to archive my images in 16bit ProPhoto, but have never really been convinced of the relevance to my shooting needs. I don't shoot skies in earnest, and my serious photography is studio based. This article just gave me the push I needed to stand up for my own instincts! 8 Bit AdobeRGB for me now, alongside the original RAW, for archiving. :-) >> Here are some more insights that you might find interesting. >> http://mike.russell-home.net/tmp/DanMargulis16Bit/ > > Wonderful read that, Mike. Many thanks! Wooohoooo. I feel L I B E R A T E > D! > Those 76Mb 16 bit ProPhotoRGB TIFFS are now history. :-) Hi Nigel, Yes, Dave Cardinal's article was particularly informative for me. Mike >>> Here are some more insights that you might find interesting. >>> http://mike.russell-home.net/tmp/DanMargulis16Bit/ [quoted text clipped - 6 lines] > > Yes, Dave Cardinal's article was particularly informative for me. It's an interesting read all around - my thoughts are ... (a) It's obviously an emotive point for Dan. I'm reminded or court room scenarios where the prosecution first outlines their case, and at the end of it you're left thinking "shoot the giulty basterd now" - and then when the defence has finished you think "hang on - that's very plausable too - definately two sides to that story". In this case we heard from a somewhat pissed off Dan, but haven't heard "the other side", so who knows what their response would be? (b) It's even conceded by Dan that 16 bit can produce a superior result with computer-generated graphics in wide colourspaces (which I would interpret to be something like a graduated fill (eg replacement sky) whilst in something like LAB colour? In my case I do use LAP colour a lot - I often do large gradient fills - the size of the files aren't an issue, and I've got the horsepower to crunch them. In summary it appears that in most situations 16 bit doesn't appear to offer an apparent advantage (but may under some circumstances) whereas (assuming that size isn't an issue) nor does it offer any disadvantage. Truth be known, I prefer a 16 bit workflow for my images, but am often presented with 8 bit images for processing by clients - so I work in both. It seems that the lesson to be learned from all of this is that in most situations it really doesn't make any difference? >>>> Here are some more insights that you might find interesting. >>>> http://mike.russell-home.net/tmp/DanMargulis16Bit/ [quoted text clipped - 34 lines] > It seems that the lesson to be learned from all of this is that in most > situations it really doesn't make any difference? I happen to enjoy Dan's style. He's the most entertaining and literary personalities in the Photoshop world, and if he's a bit testy about 16 bit versus 8 bit, I think it's understandable. It's true that the march of technology is providing us with an embarrassment of riches. Just as a rich man does not concern himself with whether a car costs 50 thousand euros or 100 thousand, perhaps many of us need not worry whether we work in 8 or 16 bits. This is at least part of the point you are making. OTOH, a factor of two is still a factor of two. No matter how rich I might be, I would enquire whether the 100 thousand Euro car had *any* demonstrable advantage over the 50 thousand Euro car. By the same token, I ask for evidence, in the form of an example, that editing in 16 bits would provide me with a demonstrable advantage over editing in 8 bits. I have not seen, as of April 14, 2006, a color photograph, in a "normal" color space, that demonstrates this. Even in psychedelic color spaces like ProPhoto RGB, examples are very rare, which should be the clue that it makes no difference to those of us who are content to work in sRGB, Adobe RGB, Colormatch, or Apple RGB. Got image? Anything else is hand waving that is as meaningful to a blind man as a sighted person. --- Mike Russell www.curvemeister.com > OTOH, a factor of two is still a factor of two. No matter how rich I > might be, I would enquire whether the 100 thousand Euro car had *any* > demonstrable advantage over the 50 thousand Euro car. By the same token, > I ask for evidence, in the form of an example, that editing in 16 bits > would provide me with a demonstrable advantage over editing in 8 bits. Perhaps this is a difference in human psyche? I was recently faced with the prospect of buying a wide-format printer. Canon had the W6400 - HP had their offerings - and Epson had the 7800. After much soul searching I was left with the thought that the Canon could PROBABLY do the job (I spoke with the local rep - I met with the national rep - we had sample prints done - I grilled them in diverse areas such as onsite and tech support etc) - the difference was I KNEW that the Epson could do the job - and that's what I bought. I feel the same about 8 -v- 16 bit. I THINK that 8 can probably do the job most (all?) of the time - but I KNOW 16 bit can. And for me, all other things are equal - so that's why I use it for my images. >> OTOH, a factor of two is still a factor of two. No matter how rich I >> might be, I would enquire whether the 100 thousand Euro car had *any* [quoted text clipped - 3 lines] > > Perhaps this is a difference in human psyche? I agree. If there are no images to demonstrate that 16 bit edits are superior, what other explanation is there? > I was recently faced with the prospect of buying a wide-format printer. > Canon had the W6400 - HP had their offerings - and Epson had the 7800. [quoted text clipped - 3 lines] > onsite and tech support etc) - the difference was I KNEW that the Epson > could do the job - and that's what I bought. You thought about the right things, and I believe you came up with a good decision. Would you have felt confident of your choice had you not looked at any printed images from the printers you were considering, and seen differences between them? > I feel the same about 8 -v- 16 bit. I THINK that 8 can probably do the job > most (all?) of the time - but I KNOW 16 bit can. And for me, all other > things are equal - so that's why I use it for my images. Most of us do not have the time to do our own research, and must rely on conventional wisdom and authorities we trust to make these determinations for us. Given the depths of time, the absence of any actual images that demonstrate the superiority of 16 bit editing should raise some interest among at least some people. April 14th, 2006 has all but passed, but there is a chance that April 15th, 2006 will be the day that, out of the millions of photographs taken every day, an image that edits better in 16 bits than 8 bits finally turns up. SignatureMike Russell www.curvemeister.com >> I was recently faced with the prospect of buying a wide-format printer. >> Canon had the W6400 - HP had their offerings - and Epson had the 7800. [quoted text clipped - 8 lines] > at any printed images from the printers you were considering, and seen > differences between them? Funnily enough I'd never seen an image from the one I ended up buying, but I did see some from it's predecessor - I also relied heavily on reviews - in-depth reviews. The Canon sales rep printed off one of my images using a local customers machine, but he presented it with quite a few creases in the paper - and it didn't reproduce some of the colours accurately. They kept saying things like "I'm sure these things can be tweaked" - but the message I was getting was that they really didn't understand the game, and were more interested in making a sale that selling me something that would meet my needs. > April 14th, 2006 has all but passed, but there is a chance that April > 15th, 2006 will be the day that, out of the millions of photographs taken > every day, an image that edits better in 16 bits than 8 bits finally turns > up. Yeah - what we REALLY need is 64 bit editing - that'll make a difference as noticeable as night and day ;) [re choosing between printers by (not) looking at images] > Funnily enough I'd never seen an image from the one I ended up buying, but > I did see some from it's predecessor - I also relied heavily on reviews - [quoted text clipped - 5 lines] > were more interested in making a sale that selling me something that would > meet my needs. Well, then kudos for at least making the right decision. Perhaps your leanings toward 16 bit image superiority have a similar incisive and intuitive validity. I myself must rely on more prosaic evidence. > Yeah - what we REALLY need is 64 bit editing - that'll make a difference > as noticeable as night and day ;) LOL. 32 bit cameras, anyway, may not be far away. Then we can take this discussion to a new plane. Photoshop already supports HDR images, floating point pixel representations with 23 bits of mantissa and 8 bits of exponent. The hardware is exerting its own pressure on our thinking, and the types of things we do with our images. SignatureMike Russell www.curvemeister.com > > Yeah - what we REALLY need is 64 bit editing - that'll make a difference > > as noticeable as night and day ;) > > LOL. 32 bit cameras, anyway, may not be far away. I think 32 bit cameras are quite far away. Right now, most cameras are only 12 bits, not even 14 bits or 16 bits. Why? Because a 16 bits A/D converter is too costly or too difficult to make? No, because the ANALOG sensor in front of it doesn't have a signal to noise ratio that justifies more than 12 bits. SignatureJohan W. Elzenga johan<<at>>johanfoto.nl Editor / Photographer http://www.johanfoto.nl > April 14th, 2006 has all but passed, but there is a chance that April 15th, > 2006 will be the day that, out of the millions of photographs taken every > day, an image that edits better in 16 bits than 8 bits finally turns up. Editing in 16-bit/c absolutely has benefits, always. As a simple test try editing in 1-bit/c, 2-bit/c, 3-bit/c etc. You will notice that the image quality does get improved as you increase the bit-depth. (use Image/Adjustment/Posterize to decrease the bit-depth). The benefit tends to get lesser and lesser as you increase the bit-depth. Is 8-bit/c enough then? Look at the difference of R=G=B=127 and R=G=B=128 (create such large patches side by side) in AdobeRGB, you will be able to visually differentiate them on the CRT. So 8-bit is very coarse. Then using the Curves, increase the contrast of that test graph (similarly like you increase the contrast of normal images, apply a slight s-shape using the Curves) now the the these patches are very easily differentiated visually. And as you continue to apply the basic editing steps the 8-bit/c quantization will happen at each of the steps, making the image more an more coarse (the quantization noise is added by each of the editing steps). Mr. Margulis expressly acknowledged the superiority of 16-bit editing in his tragi-comic 16-bit challenge (was it in 2002), in the rules of his challenge he insisted that he will fist convert the raw 16-bit/c image down to 8-bit/c, and then immediately back to 16-bit/c and the very reason for this was the 8-bit/c quantization errors (although he did express that with more "colorful" vocabulary). So 16-bit editing is always beneficial, even if the original data was only 8-bit/c. Simply convert the 8-bit/c image data to 16-bit/c mode and do editing there and you will often, if not always, see benefits. Depends of course on the "strength" of the editing that you do. Now, there are still two other questions: 1) does the 16-bit/c image data contain more useful/relevant image information than what the 8-bit/c version of that same image contain? 2) If there are more useful/relevant image information in the 16-bit/c image data then in what situations will this benefit bee seen in the normal photographic images? And in what way it will be seen. The current high end digital cameras (and scanners) have typically 12-bit analog-to-digital converters. The role of these ADC devices is to convert the (analog) signal from the sensor into the digital world. The bit-depth of the ADC is _not_ the (equivalent) bit-depth of signal that the imaging sensor provides. It is the dynamic range of the sensor that define the (equivalent) bit-depth of signal. This (equivalent) bit-depth is then further reduced slightly/somewhat by the nonlinearities and noise that the ADC produce. The current high end digital cameras (and scanners) have sensors that have less than 9 f/stops dynamic range (so less than 9 bits). Such 12-bit ADC will properly digitize that signal and since the dynamic range is limited by the noise the 3 to 4 lower most ADC bits will contain just noise. So, when considering the dynamic range there is not much benefit in having image data from the less than 9-bit sensor that was digitized by an 12-bit ADC and finally scaled by the acquire device (or the raw converter) to the 16-bit/c Photoshop space (that truthfully is just 15-bit/c) compared to the case where we have image data from the less than 9-bit sensor that was digitized by an 12-bit ADC and finllay scaled by the acquire device (or the raw converter) to 8-bit/c Photoshops space. The extra noise that there is in the 16-bit/c workflow is a small benefit here, it will smooth the quantization steps somewhat, however it is possible to add similar noise in the 8-bit/c workflow in the post-processing. The above however is simplification since the signal/noise ratio is not constant over the whole tonal range. In other words the (overall) dynamic range is not the only quality measure. The signal to noise ratio generally is the worst in the dark end of the range and improves all the way up to the light end. A sensor that has 9-bit dynamic range can have somewhat better signal/noise ratio (locally) in the lighter portions of the tonal range, from the midtones to highlights, could be one or even two bits more. Therefore the 16-bit/c image from high end imaging devices inherently do have better quality than what the 8-bit/c version of such images have. This is especially so when working in non-linear (gamma 1.8 or 2.2. or 2.5) RGB working-spaces where the tonal range from the midtones to highlights and severely compressed (by 2 to 3 bits). So when is this 16-bit/c quality benefit seen then? In the case where the acquire device image is accepted nearly as it is (only very weak editing is applied) then there will not be much, if at all, visible benefit on the CRT. On the flat panels some slight benefit could possible be seen since they are profiled by far mode complex and coarse ICC profiles than what the CRTs are. An ICC profile conversion can be very rude on the data, equivalent to a series of very severe editing steps and in the path to the monitor the ICC conversions are performed internally in 8-bit/c. More benefits will often be seen on prints, printer profiles are always extremely rude on the image data. However Photoshop allows to perform printer profiling internally in the 16-bit/c space so the final 8-bit/c data that is sent to the printer will be notably more more smooth compared to data that can be had from the 8-bit/c version. Mainly the 16-bit/c benefit is seen (and it is easily seen already on the CRT) after the image data has been more strongly edited. Such editing steps include strong Curves, density masking (or "opening up the shadows"), strong USM etc. With more elaborated techniques the 16-bit/c the benefits are glaringly seen, e.g in my work I shoot rather high dynamic range still objects and the detail in the shadows must be clearly visible with low noise so I shoot either 8 or 16 images, then average them and finally use strong density masking to open up the shadows. Another typical example is when I need to enhance a faint surface anomaly that is not visible for the unaided-eye, from the 16-bit/c average it can be enhanced amazingly well. Also in the common photography the 16-bit/c benefits are usually seen such situations, lesser noise in the shadow areas that have been "opened up" and in large smooth areas like the blue sky and smooth surfaces like body of the car etc, as the lack of banding and less noisier appearance. Timo Autiokari http://www.aim-dtp.net ... > Editing in 16-bit/c absolutely has benefits, always. ... > Mainly the 16-bit/c benefit is seen (and it is easily seen already on the > CRT) after the image data has been more strongly edited. ... If this is the case, it should be easy to provide an image. Remember, stick with the common working spaces - Adobe RGB, etc. No fancy gamma or wide gamut color spaces. SignatureMike Russell www.curvemeister.com > > Editing in 16-bit/c absolutely has benefits, always. > ... [quoted text clipped - 5 lines] > Remember, stick with the common working spaces - Adobe RGB, etc. No fancy > gamma or wide gamut color spaces. Color spaces (and conversions between them) are part of our work, whether we like it or not. I've seen quite a few images where the 16 bits version gives a better conversion to my Epson 7600 printer spaces (seen in soft proof) than the 8 bits version. Especially the 'matt black' ink spaces seem to benefit. The shadows block less quickly in the 16 bits conversion. That may not prove that there is a benifit in *editting* in 16 bits, but it does mean there can be a benefit in *working* in 16 bits. SignatureJohan W. Elzenga johan<<at>>johanfoto.nl Editor / Photographer http://www.johanfoto.nl >> > Editing in 16-bit/c absolutely has benefits, always. >> ... [quoted text clipped - 15 lines] > *editting* in 16 bits, but it does mean there can be a benefit in > *working* in 16 bits. Again - got image? SignatureMike Russell www.mike.russell-home.net > > Color spaces (and conversions between them) are part of our work, > > whether we like it or not. I've seen quite a few images where the 16 [quoted text clipped - 6 lines] > > Again - got image? Of course. How could I see this if I didn't have the image? SignatureJohan W. Elzenga johan<<at>>johanfoto.nl Editor / Photographer http://www.johanfoto.nl ** Waits for Mike's reply ** :) > ** Waits for Mike's reply ** :) I can't see whose post you are referring to, but don't you mean Timo's. :-? >> ** Waits for Mike's reply ** :) > > I can't see whose post you are referring to, but don't you mean Timo's. > :-? I think he's poking fun at me for responding so quickly to posts. What can I say - it's true. :-) Timo is likely to come up with some images, and my guess is they'll be interesting ones. SignatureMike Russell www.curvemeister.com > I think he's poking fun at me for responding so quickly to posts. What > can I say - it's true. :-) No no - not at all. I just thought his post would be like waving a red flag to a bull :) (plus the chap knows more than I'll ever know about the topic - damn - and just when I thought I had it all sorted in my head too :( >> ** Waits for Mike's reply ** :) > > I can't see whose post you are referring to, but don't you mean Timo's. > :-? (Unfortunately, whilst I was typing my cheeky reply, Mike was also posting - and he beat me to it - which made it a bit confusing. Sorry about that). >>> ** Waits for Mike's reply ** :) >> [quoted text clipped - 4 lines] > posting - and he beat me to it - which made it a bit confusing. Sorry > about that). Ah so! ;-) > I happen to enjoy Dan's style. He's the most entertaining and literary > personalities in the Photoshop world, and if he's a bit testy about 16 bit > versus 8 bit, I think it's understandable. 8< Yes, I forgot to add that I had a good chuckle. Very engaging stuff, not to mention courageous, especially when you look at the 'names' he is up against. I actually became quite shocked at how patronising they were to him and was in no small way reminded of the once very British attitude of some (nay most?) of our medical profession. Everything once was to be taken on trust and nothing should be questioned. It is very good to see there are still some who are steadfastly and politely prepared to stand their ground and bring these types down a peg or two. Far more liberating for technologies sake too, IMO. I may be completely wide of the mark here, but isn't there something about this in your own product, Mike? I kind of sense Wide Gamut CMYK has met with some 'resistance' too? > I actually became quite shocked at how patronising they were to him Keeping in mind though that that was only Dan's record of what took place. In a court of law they don't send the jury out for their deliberations after hearing from only the prosecution. >> I actually became quite shocked at how patronising they were to him > > Keeping in mind though that that was only Dan's record of what took place. > In a court of law they don't send the jury out for their deliberations > after hearing from only the prosecution. It is not accurate to characterize this simply as a controversy, or as an adversarial clash of authorities. If 16 bits is better, there should be some images to demonstrate this that will be plain to a 6 year old. SignatureMike Russell www.curvemeister.com >>> I actually became quite shocked at how patronising they were to him >> [quoted text clipped - 4 lines] > It is not accurate to characterize this simply as a controversy, or as an > adversarial clash of authorities. I'm not - I'm just politely suggesting that it's unfair to discount one group of experts in favour of another without appreciating that they may well have a different story to tell, especially when it's at a level that's waaaaay over my head. It's obvious to me that it's a pretty emotive topic for him - and in such situations it's pretty rare to have someone say "Actually, the other guy has been fair and reasonable at all times and has presented fair and reasonable arguments". If you've got a few spare moments do a Google on Gary North and have a read of how we're all doomed come "year 2000" - many of his arguments appeared well researched - authoritive - believeable - and no-one was able to disprove what he said. He also turned out to be totally wrong (not to mention a fruitcake). [eg: The y2k crisis is systemic. It cannot possibly be fixed. I think it will wipe out every national government in the West. Not just modify them- destroy them. I honestly think the Federal government will go under. I think the U.S.A. will break up the way the U.S.S.R. did. Call me a dreamer. Call me an optimist. That's what I think. ] From: "Pete Mitchell" <unsubscribe@hotmail.com> >>>> I actually became quite shocked at how patronising they were to him >>> [quoted text clipped - 9 lines] > well have a different story to tell, especially when it's at a level > that's waaaaay over my head. I'm suggesting that whether the emperor has clothes or not is not over anyone's head at all - if there is a difference there should be an example. > It's obvious to me that it's a pretty emotive topic for him - and in such > situations it's pretty rare to have someone say "Actually, the other guy > has been fair and reasonable at all times and has presented fair and > reasonable arguments". Stepping back from the personalities, it should be possible to come up with an image that proves, once and for all that the difference between 16 and 8 bits per channel is, say as noticeable as the difference between an uncompressed image and a jpeg. No such example exists, and whatever the emotions involved, this strikes me as the bedrock of any discussion involving image manipulation and quality. > If you've got a few spare moments do a Google on Gary North and have a > read of how we're all doomed come "year 2000" - many of his arguments [quoted text clipped - 6 lines] > U.S.S.R. did. Call me a dreamer. Call me an optimist. That's what I > think. ] I sat in on one of these presentations in 1999 or so, and it was very convincing indeed. As a member of the software community, I witnessed this from both sides, and I remain appalled at the fearmongering that went on. Your example of y2k is an intersting one that boils down to the same thing as the 8 vs 16 bit discussion. Both involve the well-known impossibility of proving the negative. It is not possible to prove that there is no advantage, ever, to editing an image in 16 bits per channel. All that can be said is that no one, after several years, has an example image to offer. Likewise, it was impossible to prove that there would be no y2k disaster until y2k had passed. --- Mike Russell www.curvemeister.com > Stepping back from the personalities, it should be possible to come up with > an image that proves, once and for all that the difference between 16 and 8 > bits per channel is, say as noticeable as the difference between an > uncompressed image and a jpeg. No such example exists, and whatever the > emotions involved, this strikes me as the bedrock of any discussion > involving image manipulation and quality. This is not exactly the can of worms I was originally opening, but similar. With regards to high bit *processing*, Roger Clark's example is rather dramatic. http://www.clarkvision.com/imagedetail/raw.converter.shadow.detail >> Stepping back from the personalities, it should be possible to come up >> with an image that proves, once and for all that the difference [quoted text clipped - 7 lines] > rather dramatic. > http://www.clarkvision.com/imagedetail/raw.converter.shadow.detail While my work is extremely busy right now, I did do a couple of experiments and will add the results to the web page in a couple of weeks (maybe sooner). I also read the Dan Margulis thing. I didn't know there was such a controversy going on! The interesting thing I did was go back to my raw image and converted it to an 8-bit tif using photoshop CS2. Then did the same stretching to show shadow detail. Guess what? Look at Figure 8 on the above page. I got the same result!!!! Actually, one might even pick the 8-bit tif as some of the darker parts are clipped to lower values giving the impression of more range. Then I did the same thing with the camera generated jpeg. Guess what? It LOOKS BETTER than the 8-bit or 16-bit tif that was photoshop processed! But neither comes close to the ImagesPlus 64-bit processing. Like I've said in other posts, photoshop does 15-bit processing and I believe that is causing some issues with the "16-bit" processed data. I've been experimenting doing more and more processing in ImagesPlus as it does scientifically correct math (64-bit processing, but I don't believe that 64 or even 32 bits are really warranted-it's more the math). I think this experiment shows that the main reason people have been having trouble with Margulis' challenge is not that 16-bits is not better, but the funny way photoshop processes data. I would use ImagesPlus for all processing but it doesn't have selections, let alone feathered selections, nor does it do color management. But mathematically, it does levels, upsizing, saturation correctly and has many great filters and image restoration tools not available at all in photoshop. So I do the major processing steps in ImagesPlus then to photoshop for minor tweaks and color management and ICC profiles for output. I did that on this image: http://www.clarkvision.com/galleries/gallery.bird/web/great.blue.herons.the.kiss .JZ3F8149.f-700.html and was amazed how simple my work flow became and that I had so little work to do in noise suppression of the background (none, when I normally have had to do a lot). I need to go back and work the image using photoshop only and show the resulting difference, which I believe will be significant. Now, on Margulis' side, I agree that many images do not need 16-bits, and you can produce good images with 8-bits. I have. In fact most of my images are 8-bit, even 8-bit jpegs! I have placed in Natures Best international competitions with 8-bit jpegs. Most of my bear images are 8-bit jpegs, and I make 16x24 inch prints from them: http://www.clarkvision.com/galleries/gallery.bear But now that I've found the differences in photoshop's 15-bit math versus real image processing, I find I can produce better images from true 16+ bit processing. Scientifically this makes sense. The argument for 16 versus 8 bit processing is well established in the scientific image processing community (of which I am a member). I don't use photoshop for any scientific image processing, and some photoshop tools do scientifically incorrect math, like addition when they should be doing a multiply (like the contrast tool). Then add the 15-bit trick, which, I think, is to avoid negative numbers, and speed processing, it compromises high signal-to-noise data like we are seeing from modern DSLRs. The fact that it hasn't become too much of an issue is that the gamma encoding scheme is really pretty good and enable a high dynamic range sufficient precision to not show artifacts very often. But some of Margulis' rants are wrong (as are some of the posts by others). The signal to noise and dynamic range discussions are generally incorrect. This is partly because people are analyzing data after going through the gamma correction and these raw converters that are doing funny things to the data as shown on my page above. The raw conversion is averaging adjacent pixels in the Bayer pattern, but the sensor/camera does not do this in the raw data, on either Canon or Nikon cameras that I have tested. True dynamic range is shown on these two pages (and greater than 11 stops is common on better DSLRs; the 1D Mark II gets 11.6 stops): Procedures for Evaluating Digital Camera Noise and Full Well Capacities; Canon 1D Mark II Analysis http://www.clarkvision.com/imagedetail/evaluation-1d2 The Signal-to-Noise of Digital Camera images and Comparison to Film http://www.clarkvision.com/imagedetail/digital.signal.to.noise Roger >> http://www.clarkvision.com/imagedetail/raw.converter.shadow.detail I updated the page to include 8-bit tif processing and 8-bit jpeg processing. The 8-bit tif photoshop processing produces almost identical results to 16-bit tif photoshop processing (really 15-bit). the 8-bit in camera generated raw showed better shadow detail than the photoshop CS2 8 or 16-bit raw converter. But ImagesPlus with better math and 64-bit processing produces much better shadow details and color in the shadows. Roger Can you believe that Google has indexed that page already! Quick question - can you point me to the imagesplus 2.5 website? I've never heard of it, and can't seem to find a link to them. > Quick question - can you point me to the imagesplus 2.5 website? I've never > heard of it, and can't seem to find a link to them. http://www.mlunsold.com/ >> I'm not - I'm just politely suggesting that it's unfair to discount one >> group of experts in favour of another without appreciating that they may [quoted text clipped - 4 lines] > anyone's head at all - if there is a difference there should be an > example. But that's still the danger when people read one article from Dan (pointing out how badly he's been treated by the others), without hearing what the others have to say. How would you feel about Dan's "poor treatment" if the likes of Bruce Fraser came back and said "I've sent the guy LOTS of images, but he keeps disqualifying them for one reason or another - did he not mention that?"? of if Thomas Knoll pops up and says "doing 50 conversions between RGB and LAB and back is meaningless unless you restart photoshop between conversions because it's smart enough to simply revert to a previously saved state rather than re-converting the image"? Dan may well be right - or he may well be wrong - there may well be thousands of photos floating around that illustrate the point - MY point is that I don't know and I'm not qualified to judge. As a member of the jury though I want to be fair - and for me that means waiting until I've heard from the defence as well as the prosecution. >>> I'm not - I'm just politely suggesting that it's unfair to discount one >>> group of experts in favour of another without appreciating that they may [quoted text clipped - 8 lines] > (pointing out how badly he's been treated by the others), without hearing > what the others have to say. Again, you appear to be couching this in terms of a verbal argument, and not one of seeing for yourself. > How would you feel about Dan's "poor treatment" if the likes of Bruce > Fraser came back and said "I've sent the guy LOTS of images, but he keeps > disqualifying them for one reason or another - did he not mention that?"? I've seen a number of examples from Bruce Fraser. Most, if not all of them, involve artificial gradients and not photographs and are therefore of zero interest to photographers. > of if Thomas Knoll pops up and says "doing 50 conversions between RGB and > LAB and back is meaningless unless you restart photoshop between > conversions because it's smart enough to simply revert to a previously > saved state rather than re-converting the image"? If you care about the truth or falsity of Dan's statement, why not try it yourself? It takes less than a minute. What Thomas Knoll, or anyone else, says is irrelevant because you have the means to recreate the experiment. > Dan may well be right - or he may well be wrong - there may well be > thousands of photos floating around that illustrate the point - MY point > is that I don't know and I'm not qualified to judge. As a member of the > jury though I want to be fair - and for me that means waiting until I've > heard from the defence as well as the prosecution. This is a poor analogy. A court proceeding is an adversarial, verbal exercise that has almost no relevance to scientific determination. I suggest that each of us is well capable of deciding, on our own, whether we can see the difference between editing in 8 versus 16 bit per channel. Take one of your images, edit it in 16 bits - then convert the original to 8 bits and repeat the same thing. Can you see the difference? So far there have been examples of photographs that do better in hibit - monochrome images, gamma 1.0 images, and images that are in highly stretched color spaces such as ProPhoto RGB. I have seen no examples of actual color images in the usual color spaces that benefit from hibit editing. I'm reminded of the fact that the patent office no longer requires models, except in one category - perpetual motion. Taking my cue from this, I'll let any further hand waving about hibit simply stand, hoping that people will recognize whether talk is backed up by images or not, and reserve my comments only when they are illustrated by actual example images. SignatureMike Russell www.curvemeister.com Mike, I think you've slipped off on a tangent - if you go back a few posts you'll see that I was only commenting on Siggy comment of "being shocked at how the others treated Dan" - with my response being (paraphrasing) "you need to hear both sides - how Dan says he's been treated by the others may well be different from the others recollection of the events". > Mike, I think you've slipped off on a tangent - if you go back a few posts > you'll see that I was only commenting on Siggy comment of "being shocked > at how the others treated Dan" - with my response being (paraphrasing) > "you need to hear both sides - how Dan says he's been treated by the > others may well be different from the others recollection of the events". When I said I was 'shocked', I did not wish to necessarily imply I had reached a 'verdict' on the substantive issue. ;-) I guess my emotional response (because I have no scientific background whatsoever in this area) was that I now realise I had been beguiled by all the hype about 16bit and wide gamut working spaces. Now that it is clear that only some images can derive *apparant* benefit from this technology at the output stage, based on my shooting needs I can now save myself some considerable workflow time and digital real-estate by sticking with 8bit 'finals' in future. As long as I have the original RAW, then I should also keep myself future-proof. Nigel > So far there have been examples of photographs that do better in hibit - > monochrome images, gamma 1.0 images, and images that are in highly stretched > color spaces such as ProPhoto RGB. I have seen no examples of actual color > images in the usual color spaces that benefit from hibit editing. "images that are in highly stretched color spaces such as ProPhoto RGB" So there are disadvantges to large color spaces? I guess that you need to use 16 bit with them. >> So far there have been examples of photographs that do better in hibit - >> monochrome images, gamma 1.0 images, and images that are in highly [quoted text clipped - 6 lines] > So there are disadvantges to large color spaces? I guess that you need to > use 16 bit with them. For me it's pretty much a moot point - I've got enough CPU grunt to make changes in both 8 or 16 bit appear virtually instantaniously - the extra size or a 16 bit is file isn't an issue as the images are written to CD when I'm finished and then deleted - the few filters that are 8 bit only aren't ones that I ever use - and general consensus is that, as you quote, there can be an issue with some aspects of large spaces (when I open some images I have no idea if, for example, I'll have to replace a sky with a graduated fill) - so, for me anyway, I'd be quite happy if Adobe were to hard-code a 16 bit working environment. For me, 16 bit is the safer option, with zero downside. >For me, 16 bit is the safer option, with zero downside. Mike made is as easy as: >This is a poor analogy. A court proceeding is an adversarial, verbal >exercise that has almost no relevance to scientific determination. >I suggest that each of us is well capable of deciding, on our own, whether >we can see the difference between editing in 8 versus 16 bit per channel. >Take one of your images, edit it in 16 bits - then convert the original to 8 >bits and repeat the same thing. Can you see the difference? and all our novices are awaiting the first one to take up Mike's reasonable challenge:-) Dave >> So far there have been examples of photographs that do better in hibit - >> monochrome images, gamma 1.0 images, and images that are in highly [quoted text clipped - 6 lines] > So there are disadvantges to large color spaces? I guess that you need to > use 16 bit with them. Yes - though examples are rare, there are images where editing ProPhoto in 16 bits gives superior results to editing in 8 bit. There's some interesting history here. ProPhoto was originally called ROMM RGB, for Reference Output Medium Metric, and was used by Kodak in its early film digitizing efforts. Later, it's name was changed to ProPhoto RGB, and Kodak has been promoting ProPhoto as a color space for manipulating film and digital images. http://www.scarse.org/docs/kodak/ProPhoto.pdf I don't advocate using ProPhoto, and I think that people who do use it will be at a slight disadvantage, but the choice is yours to make. SignatureMike Russell www.curvemeister.com > > How would you feel about Dan's "poor treatment" if the likes of Bruce > > Fraser came back and said "I've sent the guy LOTS of images, but he keeps [quoted text clipped - 3 lines] > involve artificial gradients and not photographs and are therefore of zero > interest to photographers. On page 24-25 of his RAW book, Bruce Fraser gives an example of 8 bits versus 16 bits that seems to qualify all your requirements. It's a normal photograph, and the corrections seem not to be excessive at all. Of course I do not know whether he secretly manipulated much more than he says, but if you would suggest that he did so just to make a point, you would treat Fraser the same way Dan says he's been treated. Quite frankly I don't see the point in this controversy. If you don't see a difference, then don't use 16 bits. If you do see it, do use it. No big deal. I don't need Bruce's or Dan's permission to make my workflow the way I prefer it. SignatureJohan W. Elzenga johan<<at>>johanfoto.nl Editor / Photographer http://www.johanfoto.nl >On page 24-25 of his RAW book, Bruce Fraser gives an example of 8 bits >versus 16 bits that seems to qualify all your requirements. It's a >normal photograph, and the corrections seem not to be excessive at all. >Of course I do not know whether he secretly manipulated much more than >he says, but if you would suggest that he did so just to make a point, >you would treat Fraser the same way Dan says he's been treated. I think for easy verification of his claim, the images would have to be on-line. >Quite >frankly I don't see the point in this controversy. If you don't see a >difference, then don't use 16 bits. If you do see it, do use it. No big >deal. I don't need Bruce's or Dan's permission to make my workflow the >way I prefer it. I don't think that Dan's point was about giving permission. It was about verifying the claim that 16-bit is somehow better. If 16-bit processing is good for B/W images, for computer generated images, and for extreme color space than it is good idea to avoid 8-bit programs (such a gimp) for those applications. If 8-bit/ch is good enough for anything else, there is no need to restrict yourself to programs that can handle 16-bit/ch. For my own software I simply use 32-bit floating point. And I use 16-bit mode in Photoshop. But I have to admit, that whenever there is such a discussion I am often tempted to 'prove' that 16-bit is better, and I usually fail. SignatureThat was it. Done. The faulty Monk was turned out into the desert where it could believe what it liked, including the idea that it had been hard done by. It was allowed to keep its horse, since horses were so cheap to make. -- Douglas Adams in Dirk Gently's Holistic Detective Agency ... > On page 24-25 of his RAW book, Bruce Fraser gives an example of 8 bits > versus 16 bits that seems to qualify all your requirements. It's a > normal photograph, and the corrections seem not to be excessive at all. This is the owl picture. in "Camera Raw with Adobe Photoshop CS2", pp 24-25. These images were created differently, one completely in Camera Raw, and the other exported as an 8 bit image to Photoshop, using unspecified manipulations. The original image, as far as I know, is not available. This is not enough of a comparison for me, or anyone else, to recreate what he did. > Of course I do not know whether he secretly manipulated much more than > he says, but if you would suggest that he did so just to make a point, > you would treat Fraser the same way Dan says he's been treated. I consider Bruce Fraser to be one of the great people in color correction. That said, I cannot recreate what he has done, based on the four images in the book, and I doubt anyone else here can either. > Quite > frankly I don't see the point in this controversy. If you don't see a > difference, then don't use 16 bits. If you do see it, do use it. No big > deal. I don't need Bruce's or Dan's permission to make my workflow the > way I prefer it. Here we are in 100 percent agreement. I think everyone has the capability to make these determinations, and not rely on any authority to do so, and I think you have advocated this consistently in your previous writings here. SignatureMike Russell www.curvemeister.com 8< > Quite frankly I don't see the point in this controversy. If you don't see > a > difference, then don't use 16 bits. If you do see it, do use it. No big > deal. I don't need Bruce's or Dan's permission to make my workflow the > way I prefer it. I wish it were that simple. Some of us, me included, need to be able to take a lead. I simply don't have the time (or probably even the nouse!) to make the qualitative and somewhat exhaustive comparative assessments necessary to build an ideal workflow and start being productive. We-pays-the-money-to-buy-the-books-that-pays-for-the-experiments-conducted-by-'experts'. If someone, considered as an authority, tells me the Emporer is wearing new clothes, I won't necessarily decide to go out into the street and look at him to see if he is naked. After having done all that, only to be told later by *a witness* that I had just wasted my money, is a little galling. Still, Caveat Emptor, and all that jazz. Nigel > I wish it were that simple. Some of us, me included, need to be able to > take a lead. I simply don't have the time (or probably even the nouse!) to [quoted text clipped - 6 lines] > that, only to be told later by *a witness* that I had just wasted my > money, is a little galling. Still, Caveat Emptor, and all that jazz. In that case, use 16 bits. Even if nobody has ever proven beyond any doubt that 16 bits is better than 8 bits, it certainly hasn't been suggested that it's the other way around. SignatureJohan W. Elzenga johan<<at>>johanfoto.nl Editor / Photographer http://www.johanfoto.nl > In that case, use 16 bits. Even if nobody has ever proven beyond any > doubt that 16 bits is better than 8 bits, it certainly hasn't been > suggested that it's the other way around. That's been my take on it as well. If one suspects that 8 bit would do the job, but one KNOWS that 16 will do it - and the side effects of size and (possibly) speed are non-issues, why not just go with the one thats got all the bases covered? In terms of my own images it's one of those "set and forget" type things. >>In that case, use 16 bits. Even if nobody has ever proven beyond any >>doubt that 16 bits is better than 8 bits, it certainly hasn't been [quoted text clipped - 5 lines] > the bases covered? In terms of my own images it's one of those "set and > forget" type things. Well, here is a counter example: Compare Figures 8, 13 and 14 at: http://www.clarkvision.com/imagedetail/raw.converter.shadow.detail The in camera generated jpeg shows more shadow detail than the 8 or 16-bit tif generated files using Photoshop CS2 (from ACR raw conversion). The jed also shows better colors in the shadows than the Rawshooter essentials 2005 16-bit generated tif. Roger ... > http://www.clarkvision.com/imagedetail/raw.converter.shadow.detail ... Roger, Thanks for adding some new information to the mix. This does not happen often. and with example images! SignatureMike Russell www.curvemeister.com > > That's been my take on it as well. If one suspects that 8 bit would do the > > job, but one KNOWS that 16 will do it - and the side effects of size and [quoted text clipped - 9 lines] > raw conversion). The jed also shows better colors in the > shadows than the Rawshooter essentials 2005 16-bit generated tif. That may well be, but that is not an 8 bits versus 16 bits issue. It's a Canon internal software versus Adobe ACR or Raw Shooters issue. SignatureJohan W. Elzenga johan<<at>>johanfoto.nl Editor / Photographer http://www.johanfoto.nl >>>That's been my take on it as well. If one suspects that 8 bit would do the >>>job, but one KNOWS that 16 will do it - and the side effects of size and [quoted text clipped - 12 lines] > That may well be, but that is not an 8 bits versus 16 bits issue. It's a > Canon internal software versus Adobe ACR or Raw Shooters issue. Yes, and that is the point. But one has to wonder why ACR or RawShooter can't at least equal an 8-bit jpeg. It seriously calls into question the effectiveness of the software, and has implications for the 8 versus 16-bit discussion that has become quite controversial. Roger > >>The in camera generated jpeg shows more shadow detail than the > >>8 or 16-bit tif generated files using Photoshop CS2 (from ACR [quoted text clipped - 9 lines] > has implications for the 8 versus 16-bit discussion that > has become quite controversial. No, I don't think it has any implications for the 8 bits/16 bits discussion, unless Canon's own Digital Photo Professional would not get the same amount of detail out of its RAW files. As long as that has not been proven, all it means is that Canon knows its own RAW files better than Adobe or Pixmantics does, which comes as no surprise to me. SignatureJohan W. Elzenga johan<<at>>johanfoto.nl Editor / Photographer http://www.johanfoto.nl >>>>The in camera generated jpeg shows more shadow detail than the >>>>8 or 16-bit tif generated files using Photoshop CS2 (from ACR [quoted text clipped - 15 lines] > been proven, all it means is that Canon knows its own RAW files better > than Adobe or Pixmantics does, which comes as no surprise to me. I do believe that it does have implications for the 8 versus 16 bit discussion. There has been a challenge to produce a 16-bit flow that shows images clearly better than an 8-bit work flow. It hasn't been done. Perhaps the reason is that the 16-bit software really isn't getting and maintaining the information, thus limiting the test. That is what the data on that page shows. Did you read the whole page? People are trying to use specific software to get certain processing but if that software has hidden issues, the results and comparisons may not be what they believe. Roger >>>>> The in camera generated jpeg shows more shadow detail than the >>>>> 8 or 16-bit tif generated files using Photoshop CS2 (from ACR [quoted text clipped - 26 lines] > certain processing but if that software has hidden issues, the results > and comparisons may not be what they believe. Johan has a point. The camera's internal processing probably uses 32 bit or whatever it is to convert the raw file. The question of 8-vs-16 is about additional processing after getting it out of raw. Do we even know how ACR processes? Not the same as core photoshop probably. Once the raw converter is done it doesn't matter wheter it's output in 8 or 16 unless you plan to more heavy stretching. >I do believe that it does have implications for the 8 versus 16 bit >discussion. There has been a challenge to produce a 16-bit flow [quoted text clipped - 5 lines] >certain processing but if that software has hidden issues, the results >and comparisons may not be what they believe. That doesn't make sense. The discussion started when somebody claimed that 16-bit was so much better in the average case, that it would be 'amateurish' to use 8-bit processing. Now obviously, you would expect those people to have some software that actually shows those benefits. Apparently, the benefits of 16-bit processing have been shown for gray scale and computer generated images, and for large gamut spaces such as ProPhoto RGB. So, to demonstrate that Adobe Photoshop makes a mess of things, you only have to put a 16-bit/ch file online together with some instructions for processing that demonstrate the effects. That different raw converters give different results should not surprise anybody. SignatureThat was it. Done. The faulty Monk was turned out into the desert where it could believe what it liked, including the idea that it had been hard done by. It was allowed to keep its horse, since horses were so cheap to make. -- Douglas Adams in Dirk Gently's Holistic Detective Agency > There has been a challenge to produce a 16-bit flow > that shows images clearly better than an 8-bit work flow. If you refer to the "challenge" laid down by Mr. Margulis some years ago then the above is not correct. Mr. Margulis insisted himself that he will first do conversion to 8-bit/c that is immediately followed by conversion to 16-bit/c, in order to avoid the cumulating 8-bit/c quantization errors. So his challenge is not at all about 8-bit/c versus 16-bit/c _workflow_, by the above requirement in his challenge he indeed agrees that the 16-bit/c workflow is superior, as it definitely is. He is only challenging the capture capabilities of the acquire devices. > Perhaps the reason is that the 16-bit software really > isn't getting and maintaining the information, thus limiting > the test. That is what the data on that page shows. That is not what your page perhaps shows. Photoshop has a range of 15bits + 1 level (so 32769 levels alltogether, from level 0 to level 32768) and it does maintain what ever image information you put into that range. But with the curent high end digital cameras you only get 9 bits (or less) of (useful) image data (even if the analog to digital converter in them is 12-bit or 14 bit deep) so you are correct in that Photoshop is not getting such data that would show benefit in the above mentioned challenge. > People are trying to use specific software to get certain processing > but if that software has hidden issues, the results and comparisons > may not be what they believe. That is indeed so and Photoshop truly has several hidden (or not well known) issues. But not maintaining the image information in the higher bit-depth mode is not among them. Also the internal processing in the digital cameras have hidden issues and even the raw camera data is rather strongly processed data, people who are trying e.g. to measure the dynamic range of these devices get results and comparisons that are not what they believe them to be. Timo Autiokari http://www.aim-dtp.net >>Perhaps the reason is that the 16-bit software really >>isn't getting and maintaining the information, thus limiting [quoted text clipped - 4 lines] > 32768) and it does maintain what ever image information you put into > that range. I disagree with this. For my work I do scientific imaging (NASA aircraft and spacecraft) and work with true 16, 32, and 64-bit processing systems. I have processed some of my photographs in 64-bit systems and do see a difference with what photoshop produces. After a number of corrections in photoshop, the images begin to take on a "pasty" look. The result is subtle but obvious the more work I do in real high bit systems. For example, see Figure 7 on this page: http://www.clarkvision.com/imagedetail/image-restoration1 and examine the fox's fur to the upper left of the eye. Compare the left most image of the figure (not done in photoshop) with the middle (photoshop). Note that the fine detail in the mid tones are smooth in the photoshop version. With noisy data this pasty look is hidden, but with high signal-to-noise ratio data from DSLRs, the problem is showing. > But with the curent high end digital cameras you only get 9 bits (or > less) of (useful) image data (even if the analog to digital converter > in them is 12-bit or 14 bit deep) so you are correct in that Photoshop > is not getting such data that would show benefit in the above mentioned > challenge. Perhaps with some cameras, but not the ones I have tested or those tested and used by other astronomers. The Canon 1D Mark II has a measured dynamic range of 3190 = 11.6 stops at ISO 100. See Table 1, column G at: Procedures for Evaluating Digital Camera Noise and Full Well Capacities; Canon 1D Mark II Analysis http://www.clarkvision.com/imagedetail/evaluation-1d2 This page shows data and references for other cameras: http://www.clarkvision.com/imagedetail/digital.signal.to.noise This page shows an image where the intensity ranges over 10 stops within a single scene; see Figure 6b: Dynamic Range of an Image http://www.clarkvision.com/imagedetail/dynamicrange Another pages with test data that shows over 10 stops of dynamic range: Dynamic Range and Transfer Functions of Digital Images and Comparison to Film http://www.clarkvision.com/imagedetail/dynamicrange2 >>People are trying to use specific software to get certain processing >>but if that software has hidden issues, the results and comparisons [quoted text clipped - 8 lines] > who are trying e.g. to measure the dynamic range of these devices get > results and comparisons that are not what they believe them to be. I agree with the issues of measuring dynamic range, and signal-to-noise ratios. Most people are doing so on raw converted data that has been massaged by the raw converter they used. But you can extract the true raw data before any Bayer interpolation artifacts are introduced with the correct software. The Canon raw data is truly raw. Nikon raw data appears to be raw except it has been quantized to something like 663 levels (I forget the exact number), but it is still linear and representative of the sensor noise and linearity. Roger I took the liberty of changing the subject line since the content has changed. > For example, see Figure 7 on this page: > http://www.clarkvision.com/imagedetail/image-restoration1 So, you present the above as a proof that the Photoshop 15-bit/c space is somehow faulty. But the Photoshopped version is damaged by the use of a huge radius. Below I show how it is done properly in Photoshop, the result is equal or even better than your Adaptive Richardson-Lucey version. Step by step instructions so you can test it yourself: 1) downloaded your fox.c01.06.2004.JZ3F3731.frmraw-crop1a.b.tif, it is 16-bit data in the sadRGB profile. (it is btw a failure to store better than 8-bit/c data as sadRGB since the sadRGB has lookup table transfer function that has only 1024 points) 2) Converted it to 'CIE 1931 d65 gamma 1.0' profile using AdobeACE engine, Relative Colorimetry intent, no blackpoint compensation. 3) Scaled it up by 200%, using Bicubic interpolation. (2x interpolation is required per your page). 4) Applied USM with A=500, R=0.5 and T=0 5) Converted it to sadRGB profile using AdobeACE engine, Relative Colorimetry intent, no blackpoint compensation, no dither. 6) Converted it to 8-bit/c 7) Saved it as JPG, quality 12 8) uploaded as: http://www.aim-dtp.net/aim/temp/fox.c01.06.2004.JZ3F3731.frmraw-crop1a.b_timo.jpg Timo Autiokari ... > Step by step instructions so you can test it yourself: > [quoted text clipped - 13 lines] > 8) uploaded as: > <http://www.aim-dtp.net/aim/temp/fox.c01.06.2004.JZ3F3731.frmraw-crop1a.b_timo.jpg> Very nice, Timo! It's not that often that I see something here that I have to immediately go and experiment with. This is one of them. Thanks! SignatureMike Russell www.curvemeister.com > I took the liberty of changing the subject line since the content has > changed. [quoted text clipped - 8 lines] > Below I show how it is done properly in Photoshop, the result is equal > or even better than your Adaptive Richardson-Lucey version. I disagree that it is even equal. Its is very good compared to what most others have submitted to me. (Bill Hilton submitted one that is equal using a Bill Atkinson action that uses edge detection). In your sample, the hairs are slightly greater in width than the RL image, but pretty close. Also, look at the iris of the eye which shows a clear difference. Then a huge difference with Figure 8 which combines bot RL and unsharp mask. Photoshop unsharp mask is not the only processing that gives pasty results. I plan to do a more rigorous study but work demands too much time right now. When I do get it done, I'll post results. Roger > Step by step instructions so you can test it yourself: > [quoted text clipped - 22 lines] > > Timo Autiokari >> I took the liberty of changing the subject line since the content has >> changed. [quoted text clipped - 23 lines] > >Roger Where would you like images sent to? Thanks -Rich >I took the liberty of changing the subject line since the content has >changed. [quoted text clipped - 21 lines] >3) Scaled it up by 200%, using Bicubic interpolation. (2x interpolation >is required per your page). One thing I can't figure out. If you scale-up the image he provides in order to work on it, the pixel size on-screen is twice that if his previously scaled-up image. Therefore, it is impossible to make it as sharp. How is the image scaled (bicubic, etc, all produce the same result) so that the pixel count increases by a factor of 4 and the pixels remain the same size? Despite this problem, I got "close" to what he did using Pixel Magic. Thanks -Rich I took the liberty of changing the subject line for this since the content does not match with the original subject line. > The Canon 1D Mark II has a measured dynamic range of > 3190 = 11.6 stops at ISO 100. See Table 1, column G at: > Procedures for Evaluating Digital Camera Noise and Full > Well Capacities; Canon 1D Mark II Analysis > http://www.clarkvision.com/imagedetail/evaluation-1d2 The two image method simply does not give correct results. The blackpoint of the camera is not steady from shot to shot and it does per shot noise reduction at the sensor level, minimal differences in the dark end between the two shots will result very large errors with this technique. The dynamic range can only be evaluated from a single shot. I to use the Mark II, I have measured the DR to be about 8.5 f/stops at ISO=100 (so about 362:1). I have two Q60 color input targets under even soft illumination and I place a neutral density filter in between, half way so that the gray range of the other target is imaged through the ND filter, it is rather easy to calculate the DR from this. Timo Autiokari http://www.aim-dtp.net > > The Canon 1D Mark II has a measured dynamic range of > > 3190 = 11.6 stops at ISO 100. See Table 1, column G at: [quoted text clipped - 3 lines] > > The two image method simply does not give correct results. It's not "two image", it's more like "20 image" (or more, and this is per ISO setting). Also note that I have data that matches Clark's more or less exactly. > The blackpoint of the camera is not steady from shot to shot and it does per > shot noise reduction at the sensor level, minimal differences in the > dark end between the two shots will result very large errors with this > technique. Clark lays out his apparatus and method clearly. Did you read them? Where is this "two shot" business coming from? > The dynamic range can only be evaluated from a single shot. The exposure control via shutter and aperture (while holding scene illumination fixed) is vastly more precise and repeatable than attempting to modulate the scene. Even worse, we have the problem: > I to use the Mark II, I have measured the DR to be about 8.5 f/stops at > ISO=100 (so about 362:1). I have two Q60 color input targets under even > soft illumination and I place a neutral density filter in between, half > way so that the gray range of the other target is imaged through the ND > filter, it is rather easy to calculate the DR from this. You've probably measured the DR of your target + filter, not the camera. Even an infinite DR camera would probably end up being measured as "8.5 stops" here. I suggest you repeat Clark's experiment as he describes -- it takes an hour or so to collect the data and another to make the plots and fit some lines. What is he doing wrong? Can your model explain his data? > I took the liberty of changing the subject line for this since the > content does not match with the original subject line. [quoted text clipped - 21 lines] > Timo Autiokari > http://www.aim-dtp.net Hmm, what is wrong with this picture? Your data says 8.5 stops maximum dynamic range at ISO 100, and my data shows 11.6 stops. But you say my method produces produces large errors. The errors you cite would increase the noise and make my determination of dynamic range too low. Thus, by your argument my range of 11.6 should be even higher. Second, the two image method is not dependent on level shifts between the two images. When you subtract two images, you get two things: an average level and noise about that average. I analyze the noise, not the average difference. So your assertion is incorrect. You can prove this yourself by taking two images, adding a constant, subtract them and analyze the standard deviation from pixel to pixel. Add a different constant and analyze again. You get the same standard deviation. Third, this method is the standard method developed years ago by electronic engineers to study CCD properties, and is the preferred method in use today. If you checked the references, you would have found that. Fourth, you ignored two other test pages where data from a single image shows over 10 stops of dynamic range. Your 8.5 stops maximum is in error. Roger > Hmm, what is wrong with this picture? Actually what is wrong is your definition: "Column G = maximum signal / read noise" that you label as the "Dynamic Range". Sure, this is one of the definitions of DR and sure it is a very good definition for the DR for many other sensing systems (like of microphones, pressure sensors, temperature sensors etc) but it does not properly characterize sensors that are sensing light. Light has a particular kind of noise itself, the Poisson noise (Photon noise, Shot Noise). So, your maximum_signal / read_noise tells you only what kind of quality you will get at the proximity of the saturation level of the sensor (for the most light/bright surfaces). It does not _at all_ characterize the capture range of the digital camera. Since the Poisson noise is squarely related to the signal your column H does quite accurately estimate the capture range of the camera, you state S/N ratio of 230:1 that equals to 7.85 f/stops. In the reality, from a single shot evalation, the Mark II does a little better, about 8.5 f/stops. > Second, the two image method is not dependent on level shifts > between the two images. When you subtract two images, you get > two things: an average level and noise about that average. > I analyze the noise, not the average difference. The two shot method is only usable with acquire devices that give you true sensor data, e.g. more than 10 year old scanners. Evaluation from a single shot is always much more reliable. Yes, you do analyze the noise, but you need to factor in the characteristics of the light, either by experiment or by theory. And even when you area analyzing the noise at the proximity of the saturation level of the sensor your 11.6 f/stops can not be correct given the 12-bit analog-to-digital converter in the camera. What you see is the effect of the noise reduction that the camera does by hardware. Timo Autiokari Come on Roger, your turn. I am enjoying this thread and learning lots. This type of measured discussion and argument is a great way to learn. regards to both Roger and Timo Don from Down Under >> Hmm, what is wrong with this picture? > [quoted text clipped - 38 lines] > > Timo Autiokari If one want to measure the DR in EVs, a test chart of adjacent patterns with reflectance which are doubled/halved in values along the chart should be shot and then one could see how many stops the DSLR can actually record. However, the Exposure Value (it's again EV) chosen, i.e., the mid-gr |
Enable EMail Alerts
Start New Thread
Reply to this Message