What happens to the hit bit data when using a curve? Let's assume we have a B&W Tiff scan with 9 bits fully populated with information. In other words, we have data at all tone values from 0 to 511. When I use Curvemeister on this image, what happens to the values from 256 to 511?
Are there any differences in how the data from 256 to 511 are handled among Curvemeister, the curves in ACR, or the curves in Photoshop CS3?
greater than 8 bits and curves
-
mikemeister_admin
- Posts: 4927
- Joined: Fri Sep 20, 2013 8:29 pm
OOH... Philosophical questions! Let me...
I say "philosophical" because I don't know where you would get a 9 bit file. If there are any commonly used graphic formats that allow 9 bit "color" information I don't know about it. I doubt Photoshop knows anything about it either. (I'm open to being corrected on this.)
Now, to the practical part of the question I think you are really asking... What happens to the extra "color" when you down-convert from 16 bit (per channel - hence forth shorten) to 8 bit? I'll give a summary answer, but if you really want to learn about Photoshop color read this book:
http://www.amazon.com/Photoshop-LAB-Color-Adventures-Colorspace/dp/0321356780/ref=pd_bbs_sr_1?ie=UTF8&s=books&qid=1196869797&sr=8-1
16 bit and 8 bit are ways of defining color in finite steps. Absolute black and absolute white are the same in either definition, but the number of steps between them is different in number. Both can describe "red" just fine and it is done with a number. However, describing just off-red gets trickier depending on how "off" of red you want to get. 8 bit has fewer steps and can describe few reds than 16 bit. 16 bit has all those extra steps to define finer differences in "red".
For example, if "red" is defined from 1 to 10 you could only have ten variations of "red". If "red" is defined from 1 to 100, you could have a lot more. Keep in mind that 1 in each number system (this is just an example) are the same color. Also, 10 and 100 are the same color of red. If you convert from the wide definition to the narrow one the ends wouldn't change. Some of the reds in the middle would be easy too; 70 would be the same color as 7.
But what do you do with a red that is defined as 55 in the wide system? There is no 5.5 in the narrow system - it all integer. One thing you can do is round to either 5 or 6. You would HAVE to do that if you were talking about just one pixel/dot of information in the picture. However, if you had a whole patch of color 55 that was together, you could convert half of them to 5 and half to 6. If you did those conversions in a checkerboard pattern, you would get a better average of 55. Up close this would be pretty obvious, but if you move back far enough you wouldn't be able to see the individual chunks of color. It would blend in your vision and look like 55 had been converted to 5.5.
In short, that's what Photoshop is doing when it converts from 16 bit to 8 bit. It is mathematically averaging a wide definition of colors (16 bit) to a narrow one (8 bit). The two end colors - black and white - aren't changing. Some of the colors in the middle that line up nicely aren't changed either. If possible (and you've configured Photoshop to do it) it will dither the colors to keep the picture looking as close as possible to the way it was. ("Dither" is what was described in the above paragraph.)
Keep in mind a couple of things. One is that you may not be able to see the differences for several reason. Probably the biggest is that your monitor probably can't show more than 8 bits of color per channel. OK, if you have a good CRT monitor it may technically be able to, but you still probably can't see it. If you have a LCD monitor that is good, it is only capable of 8 bit per channel color. [That is 8 bit in RGB and not some wider colorspace.] If you have one of the LCD monitors that is following the trend of greater speed, you probably only have 6 bits of color. They fool you by also dithering the colors to make it look like it has as many colors as an 8 bit.
Your printer dithers the colors of the handful of inks that it has to get all the colors. Dithering the colors CAN work pretty darn well. However, the more colors you have to work with before you start dithering, the better. That's why you have seen "photo" printers getting more than the traditional CMYK. Some hues and their saturation levels just can't be reached with only CMYK. Your printer probably can't hit the deepest black that is defined in your file. It almost surely can't hit the same level of brightness that your monitor can as it can only get pure white from the paper itself.
So, you can see that the conversion from whatever your Photoshop file is defined as gets even more complicated when it has to convert to your printer. That's why your printer profiles have to include the paper as well as the model (or even the specific) of printer. The "whiteness" of the paper comes into the conversion calculation. So does the tested ways the different ink colors show on that paper. This is an example of the many complicated ways that Photoshop has to do color conversions. It's amazing how well it does it all.
In summary, Photoshop uses a lot of carefully studied scientific ways of averaging color during conversions to keep your picture looking as close to what you want it to look like as possible.
Note:
This is really a Photoshop issue not specifically a Curvemeister one. I don't know, but I doubt Mike is doing his own color conversions in Curvemeister. I'd bet he is using Photoshop's color conversion techniques. I sure hope so. There should be no reason to reinvent the wheel, particularly when it is such an excellent wheel.
Sidebar...
HDR is different. HDR (High Dynamic Range) photography is a technique to blow out the ends of that defined range I was talking about above. If you have a scene that can't fit in the dynamic range defined above or even defined by your camera, how do you capture it? You have to capture the scene is multiple exposure over 3 to 7 (or more) shots. Then you blend these together in a way that doesn't force the ends back down into your 16 bit and 8 bit range.
That's why Photoshop does this in a separate process that requires 32 bits of definition to cover it. (In theory, HDR doesn't have ends to it's dynamic range, but Adobe had to make some limits because computers, software, and standards don't work well without limits. So, 32 bits it is and probably way more than enough.) In this process the whole dynamic range is kept in the blended picture. Of course, you can't see that on your 8 bit monitor, printer, or anything that I know about.
That's why Tone Mapping is used to figure out what parts of the picture get crunched down (and how much) to fit into the dynamic range of 16 and 8 bit pictures. There are some other ways to getting is crunched down, but Tone Mapping is the most commonly used and probably the best for general HDR photography. This process also does the averaging and dithering that is in a normal conversion. However, it's doesn't do it as lineally as a normal conversion. That's why a lot of HDR photos take a surreal look.
Clyde
I say "philosophical" because I don't know where you would get a 9 bit file. If there are any commonly used graphic formats that allow 9 bit "color" information I don't know about it. I doubt Photoshop knows anything about it either. (I'm open to being corrected on this.)
Now, to the practical part of the question I think you are really asking... What happens to the extra "color" when you down-convert from 16 bit (per channel - hence forth shorten) to 8 bit? I'll give a summary answer, but if you really want to learn about Photoshop color read this book:
http://www.amazon.com/Photoshop-LAB-Color-Adventures-Colorspace/dp/0321356780/ref=pd_bbs_sr_1?ie=UTF8&s=books&qid=1196869797&sr=8-1
16 bit and 8 bit are ways of defining color in finite steps. Absolute black and absolute white are the same in either definition, but the number of steps between them is different in number. Both can describe "red" just fine and it is done with a number. However, describing just off-red gets trickier depending on how "off" of red you want to get. 8 bit has fewer steps and can describe few reds than 16 bit. 16 bit has all those extra steps to define finer differences in "red".
For example, if "red" is defined from 1 to 10 you could only have ten variations of "red". If "red" is defined from 1 to 100, you could have a lot more. Keep in mind that 1 in each number system (this is just an example) are the same color. Also, 10 and 100 are the same color of red. If you convert from the wide definition to the narrow one the ends wouldn't change. Some of the reds in the middle would be easy too; 70 would be the same color as 7.
But what do you do with a red that is defined as 55 in the wide system? There is no 5.5 in the narrow system - it all integer. One thing you can do is round to either 5 or 6. You would HAVE to do that if you were talking about just one pixel/dot of information in the picture. However, if you had a whole patch of color 55 that was together, you could convert half of them to 5 and half to 6. If you did those conversions in a checkerboard pattern, you would get a better average of 55. Up close this would be pretty obvious, but if you move back far enough you wouldn't be able to see the individual chunks of color. It would blend in your vision and look like 55 had been converted to 5.5.
In short, that's what Photoshop is doing when it converts from 16 bit to 8 bit. It is mathematically averaging a wide definition of colors (16 bit) to a narrow one (8 bit). The two end colors - black and white - aren't changing. Some of the colors in the middle that line up nicely aren't changed either. If possible (and you've configured Photoshop to do it) it will dither the colors to keep the picture looking as close as possible to the way it was. ("Dither" is what was described in the above paragraph.)
Keep in mind a couple of things. One is that you may not be able to see the differences for several reason. Probably the biggest is that your monitor probably can't show more than 8 bits of color per channel. OK, if you have a good CRT monitor it may technically be able to, but you still probably can't see it. If you have a LCD monitor that is good, it is only capable of 8 bit per channel color. [That is 8 bit in RGB and not some wider colorspace.] If you have one of the LCD monitors that is following the trend of greater speed, you probably only have 6 bits of color. They fool you by also dithering the colors to make it look like it has as many colors as an 8 bit.
Your printer dithers the colors of the handful of inks that it has to get all the colors. Dithering the colors CAN work pretty darn well. However, the more colors you have to work with before you start dithering, the better. That's why you have seen "photo" printers getting more than the traditional CMYK. Some hues and their saturation levels just can't be reached with only CMYK. Your printer probably can't hit the deepest black that is defined in your file. It almost surely can't hit the same level of brightness that your monitor can as it can only get pure white from the paper itself.
So, you can see that the conversion from whatever your Photoshop file is defined as gets even more complicated when it has to convert to your printer. That's why your printer profiles have to include the paper as well as the model (or even the specific) of printer. The "whiteness" of the paper comes into the conversion calculation. So does the tested ways the different ink colors show on that paper. This is an example of the many complicated ways that Photoshop has to do color conversions. It's amazing how well it does it all.
In summary, Photoshop uses a lot of carefully studied scientific ways of averaging color during conversions to keep your picture looking as close to what you want it to look like as possible.
Note:
This is really a Photoshop issue not specifically a Curvemeister one. I don't know, but I doubt Mike is doing his own color conversions in Curvemeister. I'd bet he is using Photoshop's color conversion techniques. I sure hope so. There should be no reason to reinvent the wheel, particularly when it is such an excellent wheel.
Sidebar...
HDR is different. HDR (High Dynamic Range) photography is a technique to blow out the ends of that defined range I was talking about above. If you have a scene that can't fit in the dynamic range defined above or even defined by your camera, how do you capture it? You have to capture the scene is multiple exposure over 3 to 7 (or more) shots. Then you blend these together in a way that doesn't force the ends back down into your 16 bit and 8 bit range.
That's why Photoshop does this in a separate process that requires 32 bits of definition to cover it. (In theory, HDR doesn't have ends to it's dynamic range, but Adobe had to make some limits because computers, software, and standards don't work well without limits. So, 32 bits it is and probably way more than enough.) In this process the whole dynamic range is kept in the blended picture. Of course, you can't see that on your 8 bit monitor, printer, or anything that I know about.
That's why Tone Mapping is used to figure out what parts of the picture get crunched down (and how much) to fit into the dynamic range of 16 and 8 bit pictures. There are some other ways to getting is crunched down, but Tone Mapping is the most commonly used and probably the best for general HDR photography. This process also does the averaging and dithering that is in a normal conversion. However, it's doesn't do it as lineally as a normal conversion. That's why a lot of HDR photos take a surreal look.
Clyde
-
Bulldoggie
- Posts: 22
- Joined: Sun Aug 27, 2006 11:51 pm
My apologies for not stating my question very clearly. The heart of the issue is when and where this down sample is forced. If we bring in a raw file that's a 16 bit file with significant highlight info above 255, open it in ACR, and then immediately save in 16 bit format to a psd file, my hope is no tone mapping at all is done in creating the psd. That is, all the data above 255 is still there.
Now let's consider the same example when we apply an extreme S curve using the Point Tone Curve tool in ACR. If a compression to 8 bits isn't forced, then what transformation is used to map the values above 255 from the original raw file into the 16 bit psd. If there is a force to 8 bits, it's a bit disingenuous for Adobe to call it a 16 bit file. All we really have is 8 bits in a 16 bit container. This is nice for future calculations, which may benefit from the higher precision, but we've lost data in the deal.
The same question surfaces for Curvemeister and the curves tool in CS3. If we bring in a 16 bit tiff file from a scanner that has significant highlight detail above 255, we can bypass ACR altogether. I hope that after we open the Tiff file in CS3 and then immediately save it as a 16 bit PSD, no compression takes place. Assuming there is no compression, what happens to the high bit data when we use Curvemeister or regular curves on the file? Do we wind up with an 8 bit file in a 16 bit container?
ACR's Recovery tool got all this started with me. We all know highlights that aren't there can't be recovered. Yet Recovery can take the spike on the right side of the histogram and pull it back to show real detail. There's no scale on the ACR histogram, but we have to assume the right side isn't the largest value possible in the data file. That spike can only be the high bit data that's outside the range of the histogram. When is that data stripped off?
Right now the only tool I know of in CS3 to work with this high bit data is Recovery. It's user driven which is great, but it's crude at best. This problem is only going to get worse. Depending on who you believe the dynamic range of the human eye is somewhere between 8-10 bits, but adaptation allows us to shift that 8-10 bits over roughly the 30 bit range we experience in our world. Nikon's D300 claims it captures 14 bit data. It's not unreasonable to imagine high contrast situations filling all 14 bits. The largest value would be 16,383, yet our tools don't seem to acknowledge much above 255. It seems the first step to getting a handle on this is to understand where down sampling is forced. Later comes what kinds of tools we need to see the unseeable, and map it into the seeable - as we see fit.
Now let's consider the same example when we apply an extreme S curve using the Point Tone Curve tool in ACR. If a compression to 8 bits isn't forced, then what transformation is used to map the values above 255 from the original raw file into the 16 bit psd. If there is a force to 8 bits, it's a bit disingenuous for Adobe to call it a 16 bit file. All we really have is 8 bits in a 16 bit container. This is nice for future calculations, which may benefit from the higher precision, but we've lost data in the deal.
The same question surfaces for Curvemeister and the curves tool in CS3. If we bring in a 16 bit tiff file from a scanner that has significant highlight detail above 255, we can bypass ACR altogether. I hope that after we open the Tiff file in CS3 and then immediately save it as a 16 bit PSD, no compression takes place. Assuming there is no compression, what happens to the high bit data when we use Curvemeister or regular curves on the file? Do we wind up with an 8 bit file in a 16 bit container?
ACR's Recovery tool got all this started with me. We all know highlights that aren't there can't be recovered. Yet Recovery can take the spike on the right side of the histogram and pull it back to show real detail. There's no scale on the ACR histogram, but we have to assume the right side isn't the largest value possible in the data file. That spike can only be the high bit data that's outside the range of the histogram. When is that data stripped off?
Right now the only tool I know of in CS3 to work with this high bit data is Recovery. It's user driven which is great, but it's crude at best. This problem is only going to get worse. Depending on who you believe the dynamic range of the human eye is somewhere between 8-10 bits, but adaptation allows us to shift that 8-10 bits over roughly the 30 bit range we experience in our world. Nikon's D300 claims it captures 14 bit data. It's not unreasonable to imagine high contrast situations filling all 14 bits. The largest value would be 16,383, yet our tools don't seem to acknowledge much above 255. It seems the first step to getting a handle on this is to understand where down sampling is forced. Later comes what kinds of tools we need to see the unseeable, and map it into the seeable - as we see fit.
-
mikemeister_admin
- Posts: 4927
- Joined: Fri Sep 20, 2013 8:29 pm
Ah, that does clarify things a bit. It does take a bit of a shift from color conversion theory though.
When you open a RAW file in Photoshop you are doing a bit of conversion from one system to another. Nikon and the others don't design their system to "fit" Photoshop. They make it to fit their view of the photography world. For your purposes the "numbers" in their system don't matter. I suppose that's why they don't bother to tell you what they are.
The whole point of ACR is to get their view of the world (in the file) into Photoshop's way of dealing with the photographic world. Since that's the place you turn this raw data (pun intended) into your masterpiece, it only really matters what the numbers and colors are inside Photoshop. So, ACR is designed to get every bit of that initial data into Photoshop. By running the file through ACR you are manipulating it into Photoshop's space. You are getting every bit (another pun) into the parameters that Photoshop knows and can work with.
One of the key things you are doing in ACR is to make sure the histogram has all the data inside of it. You need to make sure that the ends of the histogram stop inside the graph. If the ends of the graph are up against the edges of the graph, you are pushing the data outside the working range of Photoshop. In this case, you will loose data at the high and/or low ends!!! The left and right edges of the ACR histogram are the limits of the highlights and shadows for regular 8 or 16 bit work space.
The histogram doesn't have numbers because they don't matter. You are still looking at the RAW data from the file. Think of the edges of that histogram box as Photoshop's boundaries. Any graph line again the edge or looking like it's going outside it will truncate data in Photoshop. The numbers only matter once you get inside Photoshop. Well, if they matter at all. The key thing to do is get all the file's data inside that histogram box.
Actually, that's about all I do in ACR. I find that Curvemeister is a MUCH faster and easier way to do my color correction and editing than the many disjointed features of ACR. So, I just make sure all my data is inside that histogram box. That gets me all the available data that is in the file into Photoshop. If I do my editing right, I won't loose any of it inside. Well, if I don't want too. More on that later.
Don't move the ends of the data WAY inside the box. This will compress the color range down more than is needed. Those 16 bits define those fine shades of colors for the whole range of the histogram. The more you shove it to the middle the fewer gradations you will have in those colors. OK, it may not matter, but you seem very concerned about loosing data. It's a fine point, but a valid one. Most of the time I just click on the "Auto" word/link for Exposure in ACR. (Bridge 2) This usually puts all that data inside the ends of the box.
Exposure in ACR doesn't control the peaks of the data that may be pushing out the top. You have to play with the other tools to pull those back into line. You probably want to do this so you make sure you have all the colors you can possible have inside Photoshop.
At this point you should have all the possible data from the RAW file inside Photoshop. Now, quit worrying about or even thinking about loosing data! The point of editing is to change the picture so it's better. That changes pixels. It will probably change all the data to some degree. So what?!? You aren't editing a bunch of numbers. You are editing a PICTURE. Focus on the picture and how good you can make it. Make art you will be proud of.
Yes, there are ways to loose data during the editing process, but that may be a good thing if it makes the picture better. Luckily for us, Curvemeister helps us make the picture better. CM helps us do this editing while loosing the least possible amount of data. It makes it easy and fast too. The key to this is the Highlight and Shadow pins.
CM lets you pin the highlights and shadows on the parts of the picture that is important to you. Unlike Photoshop's curve or ACR's tools, CM lets you SEE where the max highlights and shadows are. If you want to pin the shadows at the darkest spot, you can easily find that. By finding and pinning that, you have pushed that "color" to the far left of the histogram. You have (or may have) changed the low boundary of the range of "colors" in the photo. By pinning the highlight at the brightest spot, you have defined the right edge of the histogram.
If you pin the highlights and/or shadows at some point less than the max, you will push the data between your point and the max out of the histogram. You have cut that data out of the picture. So, that almost black stuff that you want to turn to pure black is now pure black. You will not be able to get that back. Well, except for Undo. However, this is a great artistic tool to cut out stuff that isn't important to what you are trying to say. It depends on what you are trying to say.
I bring my RAW files into Photoshop in 16 bit mode and using the ProPhotoRGB colorspace. This gets the maximum amount of information in. The first thing I do then is to convert it to Lab mode. This separates the luminosity from the color channels. That allows me to do a lot of different editing techniques on the colors and "gray" separately that I won't go into here. CM don't really need this, as you can run it in Lab mode even if the file isn't. Running in Lab also gives me the very widest possible colorspace. It would be almost impossible to create a color in Photoshop that Lab couldn't describe. Hey, Lab can define colors that you can't see.
If you do this, you can see that pinning the highlights and shadows only affect the L channel. You can see how the histogram looks and changes when you do this. Pinning the colors with the Neutral pin or any other the pins affects the "a" and "b" channels. Well, mine do because I don't usually pin brightness, but only hue and/or saturation. You can see how this pinning affects the color channels in their histograms. They can spike outside the histogram box, but who cares if your picture looks like you want it too. Yes, a spike outside the box will loose data, but it probably isn't data you will ever want back.
Adjusting the shape of a curve in CM usually doesn't push data outside the box. You would have to make some pretty big curves to do that. The picture itself would start looking pretty weird long before that happens.
The basic concept of this color adjustment is a visual one and not a numeric one. That is what is important. Editing by numbers won't create great pictures. That is why there are so few numbers associated with curves and histograms - you don't need them. Oh, 16 bit and 8 bit still describe the STEPS of the range of possible colors. However, those are steps that are INSIDE the histogram box. If you keep the ends and spikes inside the box, you won't loose any data no matter if you are in 8 bit or 16 bit modes. It isn't relevant.
BTW, I'm one of those people who doesn't see any difference between 8 bit and 16 bit. After editing my pictures, I convert to 8 bit RGB using sRGB colorspace. The 8 bit gives me smaller files that work everywhere. As I say, I can't SEE any difference! If I can't SEE a difference, it doesn't matter! Hey, it's a PICTURE and pictures are all about seeing and not about numbers. I use sRGB because it is the lowest common denominator and I know that it will display or print fine just about anywhere in the world.
In summary, use ACR to squeeze all the data you can into the borders of the histogram box. Bring it into Photoshop in ProPhotoRGB and 16 bit. Edit in Curvemeister. Use the highlight and shadow pins to make sure the range YOU WANT are pushed to the edges of the box. This gives you the detail or lack thereof in the highlights and shadows that YOU WANT. Pin the Neutral or other colors to get the colors YOU WANT. Always remember that the picture is the key thing!!! Therefore, this procedure will make sure you have all the data that you NEED and no more.
There is a chance that you will loose data, but if it improves the picture - who cares? If it improves the picture, you won't ever need that data back.
When you open a RAW file in Photoshop you are doing a bit of conversion from one system to another. Nikon and the others don't design their system to "fit" Photoshop. They make it to fit their view of the photography world. For your purposes the "numbers" in their system don't matter. I suppose that's why they don't bother to tell you what they are.
The whole point of ACR is to get their view of the world (in the file) into Photoshop's way of dealing with the photographic world. Since that's the place you turn this raw data (pun intended) into your masterpiece, it only really matters what the numbers and colors are inside Photoshop. So, ACR is designed to get every bit of that initial data into Photoshop. By running the file through ACR you are manipulating it into Photoshop's space. You are getting every bit (another pun) into the parameters that Photoshop knows and can work with.
One of the key things you are doing in ACR is to make sure the histogram has all the data inside of it. You need to make sure that the ends of the histogram stop inside the graph. If the ends of the graph are up against the edges of the graph, you are pushing the data outside the working range of Photoshop. In this case, you will loose data at the high and/or low ends!!! The left and right edges of the ACR histogram are the limits of the highlights and shadows for regular 8 or 16 bit work space.
The histogram doesn't have numbers because they don't matter. You are still looking at the RAW data from the file. Think of the edges of that histogram box as Photoshop's boundaries. Any graph line again the edge or looking like it's going outside it will truncate data in Photoshop. The numbers only matter once you get inside Photoshop. Well, if they matter at all. The key thing to do is get all the file's data inside that histogram box.
Actually, that's about all I do in ACR. I find that Curvemeister is a MUCH faster and easier way to do my color correction and editing than the many disjointed features of ACR. So, I just make sure all my data is inside that histogram box. That gets me all the available data that is in the file into Photoshop. If I do my editing right, I won't loose any of it inside. Well, if I don't want too. More on that later.
Don't move the ends of the data WAY inside the box. This will compress the color range down more than is needed. Those 16 bits define those fine shades of colors for the whole range of the histogram. The more you shove it to the middle the fewer gradations you will have in those colors. OK, it may not matter, but you seem very concerned about loosing data. It's a fine point, but a valid one. Most of the time I just click on the "Auto" word/link for Exposure in ACR. (Bridge 2) This usually puts all that data inside the ends of the box.
Exposure in ACR doesn't control the peaks of the data that may be pushing out the top. You have to play with the other tools to pull those back into line. You probably want to do this so you make sure you have all the colors you can possible have inside Photoshop.
At this point you should have all the possible data from the RAW file inside Photoshop. Now, quit worrying about or even thinking about loosing data! The point of editing is to change the picture so it's better. That changes pixels. It will probably change all the data to some degree. So what?!? You aren't editing a bunch of numbers. You are editing a PICTURE. Focus on the picture and how good you can make it. Make art you will be proud of.
Yes, there are ways to loose data during the editing process, but that may be a good thing if it makes the picture better. Luckily for us, Curvemeister helps us make the picture better. CM helps us do this editing while loosing the least possible amount of data. It makes it easy and fast too. The key to this is the Highlight and Shadow pins.
CM lets you pin the highlights and shadows on the parts of the picture that is important to you. Unlike Photoshop's curve or ACR's tools, CM lets you SEE where the max highlights and shadows are. If you want to pin the shadows at the darkest spot, you can easily find that. By finding and pinning that, you have pushed that "color" to the far left of the histogram. You have (or may have) changed the low boundary of the range of "colors" in the photo. By pinning the highlight at the brightest spot, you have defined the right edge of the histogram.
If you pin the highlights and/or shadows at some point less than the max, you will push the data between your point and the max out of the histogram. You have cut that data out of the picture. So, that almost black stuff that you want to turn to pure black is now pure black. You will not be able to get that back. Well, except for Undo. However, this is a great artistic tool to cut out stuff that isn't important to what you are trying to say. It depends on what you are trying to say.
I bring my RAW files into Photoshop in 16 bit mode and using the ProPhotoRGB colorspace. This gets the maximum amount of information in. The first thing I do then is to convert it to Lab mode. This separates the luminosity from the color channels. That allows me to do a lot of different editing techniques on the colors and "gray" separately that I won't go into here. CM don't really need this, as you can run it in Lab mode even if the file isn't. Running in Lab also gives me the very widest possible colorspace. It would be almost impossible to create a color in Photoshop that Lab couldn't describe. Hey, Lab can define colors that you can't see.
If you do this, you can see that pinning the highlights and shadows only affect the L channel. You can see how the histogram looks and changes when you do this. Pinning the colors with the Neutral pin or any other the pins affects the "a" and "b" channels. Well, mine do because I don't usually pin brightness, but only hue and/or saturation. You can see how this pinning affects the color channels in their histograms. They can spike outside the histogram box, but who cares if your picture looks like you want it too. Yes, a spike outside the box will loose data, but it probably isn't data you will ever want back.
Adjusting the shape of a curve in CM usually doesn't push data outside the box. You would have to make some pretty big curves to do that. The picture itself would start looking pretty weird long before that happens.
The basic concept of this color adjustment is a visual one and not a numeric one. That is what is important. Editing by numbers won't create great pictures. That is why there are so few numbers associated with curves and histograms - you don't need them. Oh, 16 bit and 8 bit still describe the STEPS of the range of possible colors. However, those are steps that are INSIDE the histogram box. If you keep the ends and spikes inside the box, you won't loose any data no matter if you are in 8 bit or 16 bit modes. It isn't relevant.
BTW, I'm one of those people who doesn't see any difference between 8 bit and 16 bit. After editing my pictures, I convert to 8 bit RGB using sRGB colorspace. The 8 bit gives me smaller files that work everywhere. As I say, I can't SEE any difference! If I can't SEE a difference, it doesn't matter! Hey, it's a PICTURE and pictures are all about seeing and not about numbers. I use sRGB because it is the lowest common denominator and I know that it will display or print fine just about anywhere in the world.
In summary, use ACR to squeeze all the data you can into the borders of the histogram box. Bring it into Photoshop in ProPhotoRGB and 16 bit. Edit in Curvemeister. Use the highlight and shadow pins to make sure the range YOU WANT are pushed to the edges of the box. This gives you the detail or lack thereof in the highlights and shadows that YOU WANT. Pin the Neutral or other colors to get the colors YOU WANT. Always remember that the picture is the key thing!!! Therefore, this procedure will make sure you have all the data that you NEED and no more.
There is a chance that you will loose data, but if it improves the picture - who cares? If it improves the picture, you won't ever need that data back.
-
Bulldoggie
- Posts: 22
- Joined: Sun Aug 27, 2006 11:51 pm
While I appreciate the advice, I don't think I'm any closer to an answer to my question.
What happens to the high bit data when using Curvemeister? Assume a B&W Tiff 16 bit scan with 9 bits fully populated with information. We have data at all tone values from 0 to 511. When I use Curvemeister on this image, what happens to the values from 256 to 511?
What happens to the high bit data when using Curvemeister? Assume a B&W Tiff 16 bit scan with 9 bits fully populated with information. We have data at all tone values from 0 to 511. When I use Curvemeister on this image, what happens to the values from 256 to 511?
-
mikemeister_admin
- Posts: 4927
- Joined: Fri Sep 20, 2013 8:29 pm
If you don't move the ends of the curve, nothing happens. There is no data lost. Well, assuming you don't create a histogram spike that shoots out the top of the box.
If you move the ends of the curve so that the histogram hits the edges of the box, data is probably lost. You could move the ends of the data right up to the edges of the box and not loose any data. If you push it past the edge, everything pushed past will turn to pure white or black, depending on which edge.
The extra data in a 16 bit picture is NOT at the ends; it is in the middle of the data. The above happens whether the picture is 8 bits or 16 bits. The data you loose will be very close to pure black or white. For example, if you move the shadow end of the curve so that the histogram at that end pushes up against the edge and beyond all those low numbers turn to pure black. A 5 would now be 0. You would not be able to get those pixels back to 5 again without moving all the 0 pixels to 5. Of course, you may not be able to SEE any detail in the 5 pixels anyway, but that is what happens.
I think it's time for you to start experimenting and discover for yourself. Open up a RAW file in ACR and play with the setting while you watch the histogram. Do all the settings, including the different colorspaces and bit depths. Watch what really happens. When you are happy and the whole curve is inside the histogram box, open it into Photoshop. Then open up Curvemeister. Move the curves around in all the different modes. Move the ends and move the middle of the curves. Watch the histogram curves the whole time. Then convert it to 8 bit and do it again.
By doing this you will see what is happening to the data. Any time you get part of the histogram slamming up against the left, right, or top of the box, you will be in a condition that will lose data. If the histogram curve doesn't slam up against the edges, you aren't loosing data.
You need to do this experimenting to see for yourself what is going on. Then get back to making photographs.
Clyde
If you move the ends of the curve so that the histogram hits the edges of the box, data is probably lost. You could move the ends of the data right up to the edges of the box and not loose any data. If you push it past the edge, everything pushed past will turn to pure white or black, depending on which edge.
The extra data in a 16 bit picture is NOT at the ends; it is in the middle of the data. The above happens whether the picture is 8 bits or 16 bits. The data you loose will be very close to pure black or white. For example, if you move the shadow end of the curve so that the histogram at that end pushes up against the edge and beyond all those low numbers turn to pure black. A 5 would now be 0. You would not be able to get those pixels back to 5 again without moving all the 0 pixels to 5. Of course, you may not be able to SEE any detail in the 5 pixels anyway, but that is what happens.
I think it's time for you to start experimenting and discover for yourself. Open up a RAW file in ACR and play with the setting while you watch the histogram. Do all the settings, including the different colorspaces and bit depths. Watch what really happens. When you are happy and the whole curve is inside the histogram box, open it into Photoshop. Then open up Curvemeister. Move the curves around in all the different modes. Move the ends and move the middle of the curves. Watch the histogram curves the whole time. Then convert it to 8 bit and do it again.
By doing this you will see what is happening to the data. Any time you get part of the histogram slamming up against the left, right, or top of the box, you will be in a condition that will lose data. If the histogram curve doesn't slam up against the edges, you aren't loosing data.
You need to do this experimenting to see for yourself what is going on. Then get back to making photographs.
Clyde
This is a great discussion that Mike can clear up in a few minutes I'm sure.
He knows the process he is using....so to speak...
Clyde...Awesome explaination...really...
Bulldoggie. Is there a specific application you are needing all the pixel data for? If it's for printing remember that the digital conversion to analog printing will lose some of the data as well..If It's for engineering, you could test the results against each other and see what actually happens to the data. Same image one using CM one not.
My understanding from Mike in the past has been that CM does not do any destructive changes to the image unless you force it to as Clyde has described.
My bet is that as Clyde stated, when you bring it into ACR it is going to fit into that "box". If the Pixels are in the ACR box then they are in the PSD image and in the data calculations within CM or other tools that adjust the values in the pixels.
When you convert to another bit size i.e. 24 to 16 or 8 bit image then data is definately lost. The reality is that if you make a print from the image you most likely will not see a loss in the image quality. There are some Very highend applications that might see some loss, but in general you are going to get everything you put into the image back out.
I'm pretty sure Mike will chime in to the thread when he returns from his overseas travels...Please don't be frustrated with him...He really earned his trip this year... ;)
Greg
He knows the process he is using....so to speak...
Clyde...Awesome explaination...really...
Bulldoggie. Is there a specific application you are needing all the pixel data for? If it's for printing remember that the digital conversion to analog printing will lose some of the data as well..If It's for engineering, you could test the results against each other and see what actually happens to the data. Same image one using CM one not.
My understanding from Mike in the past has been that CM does not do any destructive changes to the image unless you force it to as Clyde has described.
My bet is that as Clyde stated, when you bring it into ACR it is going to fit into that "box". If the Pixels are in the ACR box then they are in the PSD image and in the data calculations within CM or other tools that adjust the values in the pixels.
When you convert to another bit size i.e. 24 to 16 or 8 bit image then data is definately lost. The reality is that if you make a print from the image you most likely will not see a loss in the image quality. There are some Very highend applications that might see some loss, but in general you are going to get everything you put into the image back out.
I'm pretty sure Mike will chime in to the thread when he returns from his overseas travels...Please don't be frustrated with him...He really earned his trip this year... ;)
Greg
Hi Guys (officially back from Asia, complete with a cold imported from Cambodia).
Bulldoggie's questions seem to return to issues bearing on the conversion from 8 to 16 bits.
Photoshop divides by 256 to convert a 16 color value to 8 bits. So, for example, the value 65535 is changed to 255. Converting the other direction is done in a similar way, with the added detail that the 8 bit value is usually duplicated in both the top and bottom portion of the 16 bit value.
BTW - Thanks, Clyde, for an excellent summary of how dithering and other color conversion phenomena.
Mike
Bulldoggie's questions seem to return to issues bearing on the conversion from 8 to 16 bits.
Photoshop divides by 256 to convert a 16 color value to 8 bits. So, for example, the value 65535 is changed to 255. Converting the other direction is done in a similar way, with the added detail that the 8 bit value is usually duplicated in both the top and bottom portion of the 16 bit value.
BTW - Thanks, Clyde, for an excellent summary of how dithering and other color conversion phenomena.
Mike
-
mikemeister_admin
- Posts: 4927
- Joined: Fri Sep 20, 2013 8:29 pm
Gee Mike, good to see you made it back--sorry you dragged a cold with you. You don't look any worse for wear--one would never know you'd been traveling:)
My understanding is that the data is still there in ACR, of course, but without using exposure/recovery sliders to pull it back into the histogram, it is not in the PSD file. Or, therefore, the CM file. Is this correct, Mike?
If Bulldoggie is asking what happens with the application of a curve in CM, well, such an adjustment could very well drive the highlights over the 255 endpoint that was determined in ACR and is being honored by both CM and PS, no? And that data is lost if the file is saved without layers?
Is the answer to this simply, 'nope'? (except in ACR where this interpretation is not set in stone)
If we bring in a raw file that's a 16 bit file with significant highlight info above 255, open it in ACR, and then immediately save in 16 bit format to a psd file, my hope is no tone mapping at all is done in creating the psd. That is, all the data above 255 is still there.
My understanding is that the data is still there in ACR, of course, but without using exposure/recovery sliders to pull it back into the histogram, it is not in the PSD file. Or, therefore, the CM file. Is this correct, Mike?
If Bulldoggie is asking what happens with the application of a curve in CM, well, such an adjustment could very well drive the highlights over the 255 endpoint that was determined in ACR and is being honored by both CM and PS, no? And that data is lost if the file is saved without layers?
Are there any differences in how the data from 256 to 511 are handled among Curvemeister, the curves in ACR, or the curves in Photoshop CS3?
Is the answer to this simply, 'nope'? (except in ACR where this interpretation is not set in stone)
Hi Gloria,
The range of 16 bit data is 0 to 65535. The point of confusion is the the curves interface, and other interfaces, use 0-255 to mean the entire range available to an image, whether it's 8 bit or 16 bit.
Highlight recovery is done using color values that are in the normal range. There are no values that are above 255 that are pulled in to recreate highlight detail. It's also not possible to generate color values above 255, though you can compress the values to all be the same value.
BTW, it's a good bet that ACR's highlight recovery works by copying data from the blue channel (which tends to be the last channel to blow) to the red and green channels, and then bumping the contrast. In Curvemeister, a similar effect is possible by using the Lizard Tail move in Lab mode, and then changing the b channel to remove any yellow cast from the highlights.
The range of 16 bit data is 0 to 65535. The point of confusion is the the curves interface, and other interfaces, use 0-255 to mean the entire range available to an image, whether it's 8 bit or 16 bit.
Highlight recovery is done using color values that are in the normal range. There are no values that are above 255 that are pulled in to recreate highlight detail. It's also not possible to generate color values above 255, though you can compress the values to all be the same value.
BTW, it's a good bet that ACR's highlight recovery works by copying data from the blue channel (which tends to be the last channel to blow) to the red and green channels, and then bumping the contrast. In Curvemeister, a similar effect is possible by using the Lizard Tail move in Lab mode, and then changing the b channel to remove any yellow cast from the highlights.
Return to “Interesting Images”
Who is online
Users browsing this forum: No registered users and 16 guests