| Author | Thread |
|
|
02/18/2004 08:48:13 PM · #1 |
between RGB/16 and RGB/8 bit channel in PS CS & 7.0
So far, I really can;t see a difference in the pics but was wondering if anyone else knew more than I do about this function.
Mucho Thanks in advance!
|
|
|
|
02/18/2004 08:59:35 PM · #2 |
Originally posted by Rooster:
between RGB/16 and RGB/8 bit channel in PS CS & 7.0
So far, I really can;t see a difference in the pics but was wondering if anyone else knew more than I do about this function.
Mucho Thanks in advance! |
From what I have read you are suppose to be able to make edits (sharpen, etc.) without losing or degrading the quality of the image (which they say normally happens when editing an 8-bit image).
|
|
|
|
02/18/2004 09:01:56 PM · #3 |
This Bruce Fraser article might help.
|
|
|
|
02/18/2004 09:04:44 PM · #4 |
Thanks guys! Here's another question; can I switch from 8 bit to 16 without hurting the pic in anyway?
|
|
|
|
02/18/2004 10:05:18 PM · #5 |
Here's something I wrote in another, unrelated thread recently:
Originally posted by dwoolridge:
Some people (on photo.net) claim to take 8-bit images, convert them to 16-bit, then add a small gaussian blur (1%) to smooth out the histogram. I can't recommend it, but you might want to try it. You should do all additional editing in 16-bit too, although a wider color space is recommended when working with 16-bit.
Since you're only starting with 8-bits (per channel) of data, the advantages of converting to 16-bit are really not that great. Any editing that would cause banding/posterization in 8-bit are also likely to exist if you edit in 16-bit and convert to 8-bit. I really haven't done intensive research/testing in this area, but I would be very interested to hear about your experience. |
If you're very curious, take an image that requires heavy editing and try these tests:
Test 1: do all editing in 8-bit/srgb
Test 2: convert to 16-bit/srgb, do all editing, convert to 8-bit/srgb
Test 3: convert to 16-bit/adobeRGB, do all editing, convert to 8-bit/srgb
Test 4: convert to 16-bit/adobeRGB, apply the 1% gaussian, do all editing, convert to 8-bit/srgb
Just do this test once for yourself. You can approximately compare the images by using Image->Apply Image using "difference". You're probably only interested in how (1) compares with each of (2),(3),(4). After the apply image, you can use Levels to see how bad things are.
Post your results! Maybe I'll try the test too.
good luck.
|
|
|
|
02/19/2004 07:56:04 AM · #6 |
While you can switch from an 8 bit to 16 bit image and get really marginally better editing, unless you are doing ridiculously extreme adjustments, the advantages are in the noise (sorry - signal processing geek humour)
It is really only worthwhile if you start with a high bit (10,12 or higher) image in the first place. Then the extra precision is a real advantage and allows adjustments while avoiding posterisation and banding.
Some people make the mistake of confusing high bit with higher dynamic range, which is actually a different thing. Think of a staircase between dark and light - the height of the staircase is the dynamic range. The bit depth is the number of steps on the staircase - a 16 bit staircase just has more steps but still goes from the same place to the same place. You can just stop in more places on the way.
Doing the 1% gaussian blur will make the histogram look smoother, but only by introducing artifical new colours in to your image. It doesn't actually improve the colour resolution of your image, it just makes the histogram look better - and frankly, I'm not taking pictures to show how great the histogram is.
Message edited by author 2004-02-19 07:58:44.
|
|
|
|
02/19/2004 09:03:13 AM · #7 |
Yeah, like I said, I can't really recommend the 16-bit to 8-bit conversion (plus 1% gaussian). However, I did find the photo.net thread(s) where this was discussed.
Benefit of converting 8 bit files to 16 bit?
Histogramatic Error in 16 bit Mode
Although I'm interested in how much, if any, practical benefit there is when doing major editing in 16-bit mode (after converting from 8-bit), it's purely out of curiosity. |
|
|
|
02/19/2004 09:08:37 AM · #8 |
Originally posted by dwoolridge:
Yeah, like I said, I can't really recommend the 16-bit to 8-bit conversion (plus 1% gaussian). However, I did find the photo.net thread(s) where this was discussed.
Benefit of converting 8 bit files to 16 bit?
Histogramatic Error in 16 bit Mode
Although I'm interested in how much, if any, practical benefit there is when doing major editing in 16-bit mode (after converting from 8-bit), it's purely out of curiosity. |
From my experience (I have a signal processing background) and also from basic maths classes at school, there is no worthwhile advantage.
It is roughly equivalent to taking two numbers to 2 decimal places as your input, doing some operations on them, and getting a result to 10 decimal places and somehow assuming that your result is more accurate. It can only ever be as accurate as your input values.
What you do gain is potentially more room to avoid rounding errors. This really comes down to how many photoshop operations would actually shift data by fractional amounts. There is a slight advantage to the 16 bit mode there, as your error accumulation would be marginally lower. However, your input data is still 8 bit, you do not gain any actual additional precision.
So there may be a marginal advantage, but only if you are doing many processing steps in succession that introduce sub 8 bit shifts in colours. Even then, the error accumulation is pretty small.
For any given 'reasonable' image, you shouldn't be doing this much adjustments directly to the image (using adjustment layers for example mostly avoids this problem, as the errors don't accumulate, they are roughly aggregated into one operation, as far as I understand it)
|
|
|
|
02/19/2004 09:20:46 AM · #9 |
Btw, I just tried this with an image with a very smooth contour (I.e., very prone to banding with over editing)
This image if you want to try it
I took it, converted one copy to 16 bit mode and kept the other in 8 bit mode.
I then applied 3 pretty extreme curves adjustments and 2 extreme level adjustments.
Looking at the histograms, you can see a noticeable difference. The question then is, is the noticeable difference actually significant.
From there, I converted both images back to 8 bit, layered them and selected difference. From the histogram there is a 0.66 mean difference per pixel value, with a standard deviation of 1.31
That means, for very extreme adjustments (far, far more than I would ever apply to an image that was any good at all) there is on average a 0.25% difference in doing 16 bit editing in this case. In the worse case the error is about half of one percent.
I can live without worrying about those sorts of differences, so I don't see any worthwhile advantage to working in 16 bit mode, unless I have a higher than 8 bit input data.
You can try these steps yourself and see how you feel about it - but you need to remember one of the main maxims of image processing - if it looks good enough - it is good enough. We don't often show someone a histogram and say 'look how cool this is!' (I realise you aren't saying we should :) )
By eye, both images looked equally the same both before and after the adjustments.
Message edited by author 2004-02-19 11:13:37.
|
|
|
|
02/19/2004 09:25:35 AM · #10 |
Reading the photo.net discussion on the adding the 1% gaussian noise, I think the poster is just confused about actual information content. There isn't any additional information, in fact he has lost some precision but makes his histogram look better.
Although, it is true that areas that suffer from banding can be somewhat improved by the addition of some gaussian noise to break up the banding patterns. This is more a function of our visual systems that like to find patterns, than any actual mathematical benefit from a colour level sense. The noise breaks down the step edges and gives the perception of a smoother image (and can be quite a useful technique to save an image with those problems, or to roughen up a mask layer that has banding due to the limitations of a 256 greyscale image.
|
|
|
|
02/19/2004 09:32:36 AM · #11 |
Originally posted by Gordon:
For any given 'reasonable' image, you shouldn't be doing this much adjustments directly to the image (using adjustment layers for example mostly avoids this problem, as the errors don't accumulate, they are roughly aggregated into one operation, as far as I understand it) |
Yeah, the math/science is not the difficult part for me, nor a point of curiosity. One could easily construct an artificial image, combine it with a small set of operations and demonstrate the advantage of going to 16-bit. I'm really interested in a common image with a common set of operations. Comparing them on-screen or at reduce-for-web-display size isn't useful either; the prints are more important to me.
I think most people take jpegs as input and want to edit the shit out of them because they can. Bit depth and rounding/quantization just don't matter to them (and they certainly won't worry about adjustment layers).
I'm just not motivated enough to play around in this area because I pretty much only shoot raw and (even if I have 8-bit input) use adjustment layers whenever possible. |
|
|
|
02/19/2004 09:51:24 AM · #12 |
Originally posted by dwoolridge:
I'm really interested in a common image with a common set of operations. Comparing them on-screen or at reduce-for-web-display size isn't useful either; the prints are more important to me. |
What I took was a common image, with a continuous tone in one channel (the blue water) which is probably about the most extreme, but still typical example. The edits I applied were probably more extreme than would typically be used, and in the end the average difference was 0.25% between the 16bit and 8bit cases.
Note that that doesn't even say the 16bit version was 1/4 of a percent more accurate, it is just a reflection of the difference in precision of the results. (accuracy and precision being different things)
That's the underlying issue here - you don't get more colour resolution from nothing, it just appears as if there is - the results aren't actually any more meaningful or better, they are just slightly different. It is only useful if you have higher quality input data (from RAW or from scanner files)
The case I tried shows to me that even in the extremes, for a normal image, the difference in minimal to be undetectable in real world usage. I suspect that high quality JPEG compression introduces more errors than this difference.
Personally, I work on RAW files, with adjustment layers, in 16 bit mode almost exclusively, because there are significant advantages to do so. But going back from 8 bit mode doesn't gain you anything.
|
|
|
|
02/19/2004 09:52:21 AM · #13 |
Originally posted by Gordon:
You can try these steps yourself and see how you feel about it - but you need to remember one of the main maxims of image processing - if it looks good enough - it is good enough. We don't often show someone a histogram and say 'look how cool this is!' (I realise you aren't saying we should :) )
By eye, both images looked equally the same both before and after the adjustments. |
Very nice photo btw.
I played around with it and find there's really no distinguishable difference at all, even above 100%. I didn't try anything extreme, just played with levels, curves, saturation. I think it confirms what I, you, and most people would expect: A slight (epsilon!) advantage in very special cases but not worth the effort.
Bruce Lindbloom has an interesting page Dan Margulis' 16-bit challenge.
Thanks. Curiosity satisfied!
Message edited by author 2004-02-19 09:54:01. |
|
|
|
02/19/2004 09:58:06 AM · #14 |
The 'challenge' is totally bogus, if you have to convert to 8 bit before working on the 16 bit data. (you only end up with 256 discrete levels back in your 16 bit file that way)
It sounds like a stupid forum discussion gone wrong :)
There are real, justifiable reasons to keep as much information as possible in the image for as long as possible. That's why high bit editing is popular, not because 16 bit editing gives you extra precision.
I really was serious when I said that I thought people were taught this in the first couple of years of school... I had it drummed in to my head by every mathematics teacher I ever had who let us use calculators.
Interestingly, but as an aside, there are editing schemes that are using floating point colour representations, to get even more precision into the calculations - just think of the fun that'll be.
Message edited by author 2004-02-19 10:00:01.
|
|
|
|
02/19/2004 09:59:29 AM · #15 |
Originally posted by Gordon:
I really was serious when I said that I thought people were taught this in the first couple of years of school... I had it drummed in to my head by every mathematics teacher I ever had who let us use calculators. |
You're preaching to the choir! |
|
|
|
02/19/2004 10:03:14 AM · #16 |
Originally posted by dwoolridge:
Originally posted by Gordon:
I really was serious when I said that I thought people were taught this in the first couple of years of school... I had it drummed in to my head by every mathematics teacher I ever had who let us use calculators. |
You're preaching to the choir! |
Well, it was worthwhile in that it made me go and prove it to myself all over again :)
|
|
|
|
02/19/2004 10:07:46 AM · #17 |
Originally posted by Gordon:
Well, it was worthwhile in that it made me go and prove it to myself all over again :) |
And I certainly appreciate the effort, as well as the gems of info (maths/precision/accuracy/etc.) you provided.
Muchas gracias! |
|
|
|
02/19/2004 02:31:04 PM · #18 |
Hey guys! Great stuff. I did the tests & I cant see much of a difference either but I do see them in the histograms. Yet another question for you brains- how can I get a TIFF or JPEG to start out at 16 bit as opposed to 8? I can do it for RAW but not the others.
|
|
|
|
02/19/2004 03:09:27 PM · #19 |
Originally posted by Rooster:
Yet another question for you brains- how can I get a TIFF or JPEG to start out at 16 bit as opposed to 8? I can do it for RAW but not the others. |
JPEG = 8bit and if your TIFF starts out 8bit you can't do anything about it.
Your raw converter should allow you to convert to 16bit (plus some colour space) tiff.
|
|
Home -
Challenges -
Community -
League -
Photos -
Cameras -
Lenses -
Learn -
Help -
Terms of Use -
Privacy -
Top ^
DPChallenge, and website content and design, Copyright © 2001-2025 Challenging Technologies, LLC.
All digital photo copyrights belong to the photographers and may not be used without permission.
Current Server Time: 08/31/2025 10:33:19 AM EDT.
