Dan Margulis Applied Color Theory - Banding

From: "Annette Murray", INTERNET:murray@dplus.net
Date: Mon, Jan 10, 2000, 10:43 PM
RE: Banding in shadows and gradations when making RGB to CMYK conversion

We are an offset printer utilizing CTP for 99 percent of our 4-c jobs. Many of the images we receive from our large client base have banding in them.

This banding appears in shadows, soft edges and gradations.

The banding is introduced to the image when the RGB to CMYK conversion is made. The banding has become even more noticeable now that we are imaging direct to plate. Adding noise and blurring helps very little.

This is very repeatable.

It happens when designers are creating composite pages in Adobe Photoshop using RGB. They make feathered selections, create gradations, use gaussian blur, etc. Once the image is flattened and then converted to CMYK banding is introduced. It is immediately visible on the screen and very apparent in the printed piece (CTP is very unforgiving!).

Does anyone have suggestions on how to alleviate this banding introduced at the RGB to CMYK conversion stage? I realize why it is happening. RGB values are being mapped to the closest CMYK values\ (of which there are fewer, thus creating bands). Right?

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Tue, Jan 11, 2000, 12:16 AM

response by Shane Steinman from Computer To Plate Pressroom:

Actually, noise can work effectively to minimize these effects, despite protestations to the contrary. It is important however, to use "smart noise" that won't introduce unwanted noise in your image highlights and other sensitive areas. You can do this with Photoshop 5.x, but if you're using an earlier version (though I can't think why) you can use Kai's Power Tools filter called Hue Protected Noise to achieve the same effect.

Alternately, and I like this idea better, you can convert the image to a 16-bit file prior to CMYK conversion. This will open up more values for the GCR or UCR conversion and should dramatically decrease the "stepping" that you're seeing in the CMYK result. Using a 16-bit colour depth should force the application to more effectively interpolate the appropriate intervening values in a gradation. The Photoshop CMYK separation setups, I believe, are stored in percentages and therefore, may not allow the converted data to be mapped in increments finer than 1%; this despite the fact that CMYK in 8-bits will accommodate steps that are 60% smaller than that.

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Mon, Jan 10, 2000, 10:56 PM

>Does anyone have suggestions on how to alleviate this banding introduced the RGB to CMYK conversion stage?

How are you converting from RGB to CMYK. If in Photoshop - what settings are you using? With some applications, I've noticed it not uncommon to get POSTERIZATION, more so than banding. Banding I associate with gradients. Posterization I associate with colors that were originally out of gamut.

>RGB values are being mapped to the closest CMYK values (of which there are fewer, thus creating bands). Right?

Right. In either case, banding or posterization, you have a separation table that isn't doing a very good job. You might try using a different separation table. One benefit to ICC profiles is that they contain a rendering intent called "perceptual" that uses a form of gamut compression designed to sacrifice saturation so that out of gamut colors each have a defined location instead of getting mapped to a small block of CMYK values; i.e., colors that are in-gamut also lose some saturation in order to "make room" for out of gamut colors.

---

From: Dan Margulis, INTERNET:76270.1033@compuserve.com
Date: Tue, Jan 11, 2000, 7:29 AM

Shane writes:

<<Alternately, and I like this idea better, you can convert the image to a 16-bit file prior to CMYK conversion. This will open up more values for the GCR or UCR conversion and should dramatically decrease the "stepping" that you're seeing in the CMYK result. Using a 16-bit colour depth should force the application to more effectively interpolate the appropriate intervening values in a gradation.>>

I like this idea better too. By converting the RGB file to 16-bit before conversion to CMYK, and reconverting it to 8-bit afterwards, we avoid a lot of the problems caused by the dot gain compensation on conversion.

<< The Photoshop CMYK separation setups, I believe, are stored in percentages and therefore, may not allow the converted data to be mapped in increments finer than 1%; this despite the fact that CMYK in 8-bits will accommodate steps that are 60% smaller than that.>>

No, everything is in 8-bit in CMYK. Photoshop just reports out the data in 1% to cater to CMYK fogies who don't understand other values.

---

From: Rick Boden, INTERNET:blphoto@cadvision.com
Date: Tue, Jan 11, 2000, 1:24 AM

Annette Murray wrote:

>It happens when designers are creating composite pages in Adobe Photoshop using RGB. They make feathered selections, create gradations, use gaussian blur, etc. Once the image is flattened and then converted to CMYK banding is introduced. It is immediately visible on the screen and very apparent in the printed piece (CTP is very unforgiving!).

I don't have a answer other than adding noise but I wanted to make a comment because I deal with banding all the time with RGB outputs. I am willing to bet that there is significant banding in the RGB file before you convert to CMYK. You may have to look at the individual channels at 100% to see it but I'm sure it is there. I don't know if trying to fix it first in RGB by adding noise would be a solution or not.

Wait, I just tried it and there is quite a difference when you add noise before converting to CMYK. Try it!

Incidentally, my solution (not applicable in your case, I don't think) for eliminating banding in images that require lots of grads and blurs is to do the work in Live Picture, a 16 bit program than all but eliminates the banding even when converted back to 8 bit per channel Photoshop.

---

From: Dan Margulis, INTERNET:76270.1033@compuserve.com
Date: Tue, Jan 11, 2000, 7:31 AM

Rick Boden writes:

<<I am willing to bet that there is significant banding in the RGB file before you convert to CMYK. You may have to look at the individual channels at 100% to see it but I'm sure it is there.>>

Looking at the individual channels won't help. Any banding that we see is likely false. Things that band on the monitor may not band on the platesetter, and vice versa. The monitor is only a 256-level per channel device just like the platesetter is. Furthermore, what you see on the monitor runs through a variety of profiles that can introduce optical jumps that aren't really there.

---

From: Art Department, INTERNET:art@scottcooper.com
Date: Tue, Jan 11, 2000, 11:42 AM

Annette Murray Wrote:

>Does anyone have suggestions on how to alleviate this banding introduced atthe >RGB to CMYK conversion stage?
I had problems with banding, and contacted Adobe. Below are their suggestions, even though your going direct to plate you might find them usefull, Good luck!

>Shane Steinman Wrote:

>It is important however, to use "smart noise" that won't introduce unwanted noise in your image highlights and other sensitive areas. You can do this with Photoshop5.x

I'm using 5.x and am not familiar with "Smart Noise" I'd like to know more. Where is it hiding?

______BELOW IS ADOBE'S RESPONSE TO THE ISSUE OF BANDING_________

Please note, if you reply to this e-mail or send an e-mail to Techdocs@adobe.com, your message will be handled by our automated e-mail back system. You will not receive a response from an Adobe technician.

Adobe Systems Technical Document 317020

Title: Banding Occurs in a Gradient or Feathered Part of a CMYK Photoshop Image

Issue

Banding (i.e., tones that are stepped or drop off abruptly) occurs in a gradient or feathered part of a CMYK image in Adobe Photoshop 5.0 and earlier.

Solutions

Do one or more of the following:

Solution 1

If you are printing to an imagesetter, increase the output resolution. For example, if the resolution is set to 2540 dpi, increase it to 3000 dpi.

For instructions, see your imagesetter's documentation.

Solution 2

Lower the halftone screen frequency (measured in lines per inch, or lpi) value of your page-layout program or output device, and then, if needed, lower the image resolution (measured in pixels per inch, or ppi).

NOTE: As a guideline, the image resolution should be twice the lpi value (ppi = 2 x lpi). In many cases, you can reduce the resolution to 1.5 times the lpi and still get acceptable results. For example, if the image resolution is 300 ppi and the lpi value is 150, try reducing the image resolution to 225 ppi. If possible, lower the lpi value (e.g., from 150 to 133), and then reduce the image resolution to 200 ppi.

For instructions to lower the lpi value, see the documentation for your page-layout program or output device, or consult your service provider.

For more information about reducing (resampling) image resolution, see the Photoshop User Guide.

Solution 3

Re-create the gradient in CMYK mode instead of creating it in RGB mode and then converting it to CMYK mode.

Solution 4

Shorten the length of the gradient, increase the number of shades (i.e.levels, values) of gray, or both:

-- Shorten the length of the gradient for blends or the feather radius for vignettes. (The longer the gradient, the more noticeable the transition between each shade of gray.)

-- Increase the shades of gray in the gradient. For example, if you create a 0-10% gradient over 7 inches, the 11 shades of gray may be too few to create a smooth transition.

Solution 5

In Photoshop, add a small amount of noise to the image by doing one of the following:

-- Add noise when creating the gradient by using the Dither option in the Gradient Tool Options dialog box.

-- Apply the Add Noise filter to the image (choose Filter > Noise > Add Noise) :

- For a CMYK image, add noise to only the affected channels and set Distribution to Gaussian for the best results. Often, banding is prominent in only one or two channels, so it's not necessary to add noise to the composite image (i.e., to all channels). Additionally, noise added to only one or two channels is not as noticeable as noise added to all channels.

The amount of noise you can apply before an image becomes grainy depends on the resolution of the image. For a 300 ppi image, for example, use a setting between 2 to 6.

- For an RGB image, apply noise to only the gray values by selecting Monochrome in the Add Noise dialog box. (Noise added to gray values is less noticeable than noise added to RGB channels.)

-- Convert a CMYK image to L*a*b color mode, then apply noise to the L (Lightness) channel.

Additional Information

Photoshop images have channels that represent color intensities as shades of gray. Each channel is an 8-bit grayscale image -- it can have up to 256 different shades of gray ranging from 0 (black) through 255 (white). RGB images have three channels; CMYK images have four. Banding occurs if an image does not contain enough shades of gray to render gradients at the existing screening settings. When printing to a PostScript printer, you can use the following formula to determine whether an image has enough gray shades for the gradient to render completely:

[output resolution/screen frequency] (squared) + 1 = The number of possible gray shades

The number of gray shades increases as you increase the output resolution, lower the screen frequency, or both. Since the scanning or image resolution is usually based on the screen frequency value (dpi or ppi = 1.5 to 2 times the lpi), reducing the resolution will also increase the number of possible gray shades. For example, if an imagesetter has a 1200 dpi resolution, and an image has a 150 lpi screen frequency, the formula is [1200/150] (squared) + 1 = 65 possible gray shades.

If you increase the imagesetter's output resolution to 2540 dpi, the formula is [2540/150] (squared) + 1 = 286 possible gray shades

Since 286 exceeds the 256 limit of possible gray shades, the resulting gradient will be much smoother than for 65 gray shades. For more information on creating smooth gradients, see Related Records. Another factor that can influence the way a gradient prints is the shape of the halftone dot. For most purposes, a round dot is ideal. But in some cases, an elliptical dot creates a superior transition between values in the midtone range. Stochastic screening (also known as frequency modulated, or FM, screening) may also improve results since the shape and the placement of dots is random rather than regular.

For more information on screening technologies, see Related Records. Often, gradients between RGB colors create values that are outside the CMYK gamut, or range. These out-of-gamut values may convert to the same in-gamut CMYK value, which may result in banding. Gradients created in CMYK mode are usually smoother because they have in-gamut values. In some cases, banding is unavoidable, so adding noise helps camouflage it. Small amounts of noise added to the affected channels are indistinguishable when printed, since a halftone dot pattern dithers the pixels. Blurring the image can also reduce banding, but will reduce the image qualiate.

------------------------------------------------------------

Information sent from Adobe via techdocs@adobe.com is property of Adobe.

Copyright (c) 1995-99 Adobe Systems Incorporated. All rights reserved.

Adobe is a trademark of Adobe Systems Incorporated. All brand or product names used in technical records are trademarks or registered trademarks of their respective holders.

---

From: Andrew Rodney, INTERNET:andrew@digitaldog.net
Date: Tue, Jan 11, 2000, 10:53 AM

on 1/11/00 8:40 AM, Chris Murphy at lists@colorremedies.com wrote:

> I'm pretty sure there is a 10-bit limit (?) for video cards on the Macintosh which would give a card with sufficient RAM 1,024 leves per channel. In any event, banding or posterization with 256 discernable levels on a monitor is a rare event.

The older Radius Thunder cards had a 10 bit DAC. It was said to produce even better results when mated to a PressView.

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Tue, Jan 11, 2000, 10:42 AM

Dan writes:

>The monitor is only a 256-level per channel device just like the platesetter is.

That's not entirely accurate. It doesn't have a levels limitation per se, but is limited to the levels of the device driving it - the video card.

I'm pretty sure there is a 10-bit limit (?) for video cards on the Macintosh which would give a card with sufficient RAM 1,024 leves per channel. In any event, banding or posterization with 256 discernable levels on a monitor is a rare event. Perhaps Michael Stokes would care to comment?

> Furthermore, what you see on the monitor runs through a variety of profiles that can introduce optical jumps that aren't really there.

All of those go through 16-bit conversion anyway, so that's not the problem. What you see on the monitor is going through ONE more conversion than what you had with Photoshop 4. As long as it's well behaved, that's not the cause of the problem. This feature is also easy to disable if you think you're seeing posterization/banding.

RGB-CMYK, CMYK-RGB (not sure about grayscale, but I think they are also) conversions are all performed through 16-bit precision. The difference between converting FIRST to 16-bit mode, then converting to CMYK, then converting to 8-bit mode - rather than just doing a mode change, is that the former adds some noise for you. When Photoshop performs a mode change from 8-bit RGB to 8-bit CMYK (using 16-bit precision) it doesn't add noise; but it does add noise when converting from 16-bit to 8-bit within the same color model.

This noise function was added in Photoshop 5.

---

From: Rick Boden, INTERNET:blphoto@cadvision.com
Date: Tue, Jan 11, 2000, 10:31 AM

Dan Margulis wrote:

>Looking at the individual channels won't help. Any banding that we see is likely false. Things that band on the monitor may not band on the platesetter, and vice versa. The monitor is only a 256-level per channel device just like the platesetter is. Furthermore, what you see on the monitor runs through a variety of profiles that can introduce optical jumps that aren't really there.

I'm having some trouble with this. Looking at individual channels just makes it easier to see but it is visible in full colour mode as well. The banding I see on my monitor at 100% looks as real as any other subtle effect that does get reproduced. Why is only the banding false?

---

From: Michael Stokes, INTERNET:mistokes@Exchange.Microsoft.com
Date: Tue, Jan 11, 2000, 1:11 PM

> That's not entirely accurate. It doesn't have a levels limitation per se, but is limited to the levels of the device driving it - the video card.

> I'm pretty sure there is a 10-bit limit (?) for video cards on the Macintosh which would give a card with sufficient RAM 1,024 leves per channel. In any event, banding or posterization with 256 discernable levels on a monitor is a rare event. Perhaps Michael Stokes would care to comment?

OK. If I'm understanding this correctly, then in the desktop printing world, this is called contouring. It is caused when you don't have enough digital counts to provide a smooth gradation change in color, so you get quantized steps of color. We use banding to describe a mechanical artifacts from the print engine itself, like inkjets that don't dry well before the next swatch of ink is laid down.

Your problem, in my opinion, has little or nothing to do with your display and everything to do with your application that is converting from RGB to CMYK. The suggestions of converting first to 16bit and then changing color spaces (and possibly adding noise in 16bit) is a reasonably one. I'm afraid you are running into some subtleties of Photoshop (ironically the same ones that led to the misleading claims about sRGB gamut being much smaller than CMYK or other display RGB spaces.) You could also try a different application just to isolate the problem. Everybody uses different algorithms so maybe Quark, Linocolor or even PageMaker will yield different results. If the banding changes (gets worse or better) with only a change in application, you have successfully identified the culprit. At a minimum, this will provide more backing when you talk with the vendor.

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Tue, Jan 11, 2000, 2:02 PM

From Stephen Trentmann, CTPP list:

Annette,
I know what you are going through. Optronics screening put me through the same hell.

Where is your screening being done? We were using a Harlequin rip but were forced to use Optronics screening on an Aurora 025. Banding showed up in everything!!!

We switched to a Harlequin based NT rip using "Harlequin" precision screening with the gray levels set higher than 1000. However using Round Square Round dot shapes, banding still showed up occasionally. I then switched to the Rhomboid (diamond) dot shape and adjusted my angles slightly,
Y = 60
M = 75
C = 15
K = 45.

Banding is almost completely gone. If it does show on the screen before proofing, the only thing I have done to reduce it is to add noise, 1 pixel at a time, two times.
Hope this helps,
Stephen Trentmann
Prepress Manager
UV Color
2430 Prior Ave. N.
Roseville, MN 55113
1-800-626-5676

---

From: Dan Margulis, INTERNET:76270.1033@compuserve.com
Date: Tue, Jan 11, 2000, 4:24 PM

Annette writes:

<<We are an offset printer utilizing CTP for 99 percent of our 4-c jobs. Many of the images we receive from our large client base have banding in them. This banding appears in shadows, soft edges and gradations. The banding is introduced to the image when the RGB to CMYK conversion is made. The banding has become even more noticeable now that we are imaging direct to plate. Adding noise and blurring helps very little.>>

If you are having more problems with CTP than previously, the chances are there is a resolution issue with your platesetter. If you're using a 133 line screen you really need 2400+ spots per inch platesetter resolution, and more if the screen is finer.

<<Does anyone have suggestions on how to alleviate this banding introduced at the RGB to CMYK conversion stage? I realize why it is happening. RGB values are being mapped to the closest CMYK values (of which there are fewer, thus creating bands). Right?>>

Not quite. There are the same number of values in CMYK, but because things get allocated differently certain ones can't be used. An oversimplified look at the process: you start with a smooth gradient that has 256 different values, which is all you can have in a given channel. Now you do some conversion, such as preparing it for some kind of output device. Say that the output device naturally prints dark. You have to convert the original gradient to make it lighter. The normal way of doing this is to apply some kind of curve that lightens the midpoint and tapers off the lightening as we approach the white and black point.

Lightening in this way will force certain points that used to be in the darker half of the image, into the lighter half. So, instead of having 128 values in the lighter half and 128 in the darker half, we may have, say, 150 in the lighter half and 106 in the darker half.

Except there are only 128 holes in the lighter half. Therefore, some of the values will have to share. Certain values that used to be different now have to be the same. Meanwhile in the darker half, there are 128 holes, but only 106 candidates to fill them. So, certain holes won't be filled. The gradient will be less smooth, because every now and then there will be a jump of two values, rather than one. The lightening changed the gradient from having 256 values in 256 spaces, to 234 values in 256 spaces.

This type of thing can happen
*when changing colorspaces
*when changing flavors of RGB or CMYK
*when sending to an output device that has some sort of calibration routine
*when applying any type of correction to the gradient.

It also can happen, deceptively, on your screen. If you are looking at a CMYK file especially, although it does happen in other colorspaces too, the calibration of your monitor may produce what looks like banding, but isn't.

Remember, the monitor only has 256 possible values per channel also.

The general solution for banding is to add noise, but there are some case (like yours) that the banding is too severe for the noise to wipe it out altogether.

The likely suspects for causing the banding in your case, with solutions, in roughly the order of likelihood:

1) The dot gain compensation that is an integral part of the RGB>CMYK conversion. This applies lightening to the file, hence may cause banding as described above. Solution: convert the RGB file to 16-bit before converting, reconvert to 8-bit after it's in CMYK.

2) The platesetter is contributing in either or both of two ways: it may have some kind of calibration routine that lightens or darkens incoming files; and/or it may lack enough resolution to construct 256 levels of gray dots. Solution: fix these things.

3) If parts of the gradient are near the edge of the CMYK gamut, there may be some smashing together of colors as part of the separation algorithm, and this may cause banding. Solution: after the file is in CMYK, Gaussian Blur Radius 4 pixels, then Filter: Fade>Color.

---

From: INTERNET:shane@archangel.net, INTERNET:shane@archangel.net
Date: Tue, Jan 11, 2000, 6:08 PM

Hi Annette,

Well, I tried my own advice and converted an RGB gradient to 16-bit colour depth before going to CMYK, but achieved only minimal success.

By far, the best approach is to add noise to the blend or gradient. It matters little whether you add the noise before or after CMYK conversion.

Use the KPT Hue Protected Noise filter for best results.

This will keep your white areas clean. You may also wish to select the problem areas with a feathered selection to minimize the transition from "noised" to "un-noised" areas of the image.

By the way, is there any reason that the artist can't work in CMYK in the first place? This would take a lot of pressure off the back end of the process and minimize the need for handling most blends or grads with kid gloves.

Shane Steinman
Project Head, dMACS Initiative Digital Magazine Advertising Canadian Specifications c/o ArchAngel Media Inc.
15 Kirkland Court, Richmond Hill, ON, CANADA L4C 9H4
Tel/Fax: (416) 410-3355 E-mail: shane@dMACS.org
Website: www.dMACS.org

---

From: Rob Mackert, INTERNET:rmackert@siennaimaging.com
Date: Tue, Jan 11, 2000, 5:28 PM

Dan,
I work for the Digital Camera division for Sienna Imaging. I run into this banding issue often and the photographers always blame the camera. You explained this in terms most people can understand. Would you mind if I used it with some of our clients?

Rob Mackert
Applications Specialist
Sienna Imaging
612-829-5444
rmackert@siennaimaging.com

---

From: Dan Margulis <76270.1033@compuserve.com>
Date: Tue, 11 Jan 2000 16:21:11 -0500

Annette writes:

> <<We are an offset printer utilizing CTP for 99 percent of our 4-c jobs.

> Many of the images we receive from our large client base have banding in them.

> This banding appears in shadows, soft edges and gradations.

> The banding is introduced to the image when the RGB to CMYK conversion is made. The banding has become even more noticeable now that we are imaging direct to plate. Adding noise and blurring helps very little.>>

> If you are having more problems with CTP than previously, the chances are there is a resolution issue with your platesetter. If you're using a 133 line screen you really need 2400+ spots per inch platesetter resolution, and more if the screen is finer.

> <<Does anyone have suggestions on how to alleviate this banding introduced at the RGB to CMYK conversion stage? I realize why it is happening. RGB values are being mapped to the closest CMYK values (of which there are fewer, thus creating bands). Right?>>

> Not quite. There are the same number of values in CMYK, but because things get allocated differently certain ones can't be used. An oversimplified look at the process: you start with a smooth gradient that has 256 different values, which is all you can have in a given channel. Now you do some conversion, such as preparing it for some kind of output device. Say that the output device naturally prints dark. You have to convert the original gradient to make it lighter. The normal way of doing this is to apply some kind of curve that lightens the midpoint and tapers off the lightening as we approach the white and black point.

> Lightening in this way will force certain points that used to be in the darker half of the image, into the lighter half. So, instead of having 128 values in the lighter half and 128 in the darker half, we may have, say, 150 in the lighter half and 106 in the darker half.

> Except there are only 128 holes in the lighter half. Therefore, some of the values will have to share. Certain values that used to be different now have to be the same. Meanwhile in the darker half, there are 128 holes, but only 106 candidates to fill them. So, certain holes won't be filled. The gradient will be less smooth, because every now and then there will be a jump of two values, rather than one. The lightening changed the gradient from having 256 values in 256 spaces, to 234 values in 256 spaces.

> This type of thing can happen
*when changing colorspaces
*when changing flavors of RGB or CMYK
*when sending to an output device that has some sort of calibration routine
*when applying any type of correction to the gradient.

> It also can happen, deceptively, on your screen. If you are looking at a CMYK file especially, although it does happen in other colorspaces too, the calibration of your monitor may produce what looks like banding, but isn't.

> Remember, the monitor only has 256 possible values per channel also.

> The general solution for banding is to add noise, but there are some case (like yours) that the banding is too severe for the noise to wipe it out altogether.

> The likely suspects for causing the banding in your case, with solutions, in roughly the order of likelihood:
> 1) The dot gain compensation that is an integral part of the RGB>CMYK conversion. This applies lightening to the file, hence may cause banding as described above. Solution: convert the RGB file to 16-bit before converting, reconvert to 8-bit after it's in CMYK.
> 2) The platesetter is contributing in either or both of two ways: it may have some kind of calibration routine that lightens or darkens incoming files; and/or it may lack enough resolution to construct 256 levels of gray dots. Solution: fix these things.
> 3) If parts of the gradient are near the edge of the CMYK gamut, there may be some smashing together of colors as part of the separation algorithm, and this may cause banding. Solution: after the file is in CMYK, Gaussian Blur Radius 4 pixels, then Filter: Fade>Color.

---

From: "Annette Murray", INTERNET:murray@dplus.net
Date: Tue, Jan 11, 2000, 5:22 PM

Thanks to everyone for their help and advice.

> If you are having more problems with CTP than previously, the chances are there is a resolution issue with your platesetter. If you're using a 133 line screen you really need 2400+ spots per inch platesetter resolution, and more if the screen is finer.

CTP is NOT introducing any additional banding. What we see on our digital proof (Polarproof) and what we see on our monitors (BARCO) is what we get on press.

I am going to try several of the suggestions listed in these many posts:

1. Adding noise before converting to CMYK (although at this point there isn't any banding).
2. Trying several different CMYK to RGB conversions.
3. One and three below.

> 1) The dot gain compensation that is an integral part of the RGB>CMYK conversion. This applies lightening to the file, hence may cause banding as described above. Solution: convert the RGB file to 16-bit before converting, reconvert to 8-bit after it's in CMYK.

> 3) If parts of the gradient are near the edge of the CMYK gamut, there may be some smashing together of colors as part of the separation algorithm, and this may cause banding. Solution: after the file is in CMYK, Gaussian Blur Radius 4 pixels, then Filter: Fade>Color.

---

From: Dan Margulis, INTERNET:76270.1033@compuserve.com
Date: Tue, Jan 11, 2000, 7:34 PM

Rick writes:

<<I'm having some trouble with this. Looking at individual channels just makes it easier to see but it is visible in full colour mode as well. The banding I see on my monitor at 100% looks as real as any other subtle effect that does get reproduced. Why is only the banding false?>>

Because artificial gradients are the only type of image that potentially have a problem if reproduction is off even by a single value. Only in a gradient will you find a significantly large area that varies from its neighbor by *exactly* one value. In a photographic image, one would never find this. In a photographic image, the same monitor problem occurs (that is, certain one-value differences display as two-values) but it's impossible to detect. In a gradient, it can be very evident, and show up as banding.

A monitor and an imagesetter each normally are capable of 256 values per channel. The same principle that causes banding in one may cause it in the other, but typically the banding will show up in slightly different areas. We are *much* more likely to see false banding in a CMYK file, because in addition to the general problem of monitor behavior, there's a second conversion thrown in: Photoshop internally converts the CMYK file to RGB in order to show it to us on screen.

If you doubt this, open up a CMYK file that displays banding, and look at the worst channel. Now, change the dot gain number in CMYK Setup. That won't change the file--but watch the banding change position. Or, you can demonstrate it in either CMYK or RGB if you have a copy of the old Gamma utility around. Change the Gamma setting, and watch the banding appear and reappear, even though the file itself doesn't change.

Maybe the most convincing way is to open up a channel that shows mild banding on screen, and blur it heavily. Obviously, this will reduce the banding. However, the screen display will probably show *increased* banding in the channel, though probably not in the composite image.

---

From: Lee Varis, INTERNET:varis@varis.com
Date: Tue, Jan 11, 2000, 10:11 PM

I just thought I might mention a good noise trick while the subject of banding fixes is still warm. This may be the kind of "smart noise" that Shane has refered to and is IMHO the best way to apply noise in any situation, RGB or CMYK.

1. Start by making a new layer in Photoshop, select Overlay mode and check the "fill with overlay neutral color" radio button in the dialog. You can also use Soft Light for a more subtle application of noise or Hard Light for heavier noise.

2. Apply noise to this neutral gray layer only. In RGB this will be a true 50% gray but in CMYK the channel values depend on the CMYK set-up.

3. Flatten before output

This method has the advantage of ramping the noise off in highlights and shadows with zero noise in black and white values thus preventing a dirty look. This also allows you to add more noise that you might be able to otherwise in the most troublesome areas of banding. You might also experiment with applying slightly different amounts of noise to individual channels (still only on your noise layer) with higher percentages in channels with the most banding.

Be carefull with the KPT noise filters. The only noise filters that work properly were in version 1.0 of Kais Power Tools. In all subsequent versions they attempted to speed up the filters and ruined them the noise tesselates in various harringbone type patterns depending on amount. Version 1.0 filters only came in 3 flavors minimum, medium & maximum. These filters are still great for noise in alpha channels where you can't use the layer noise trick easily.
Good Luck !

Best Regards,
Lee Varis
voice: (323) 937-3793
fax: (323) 937-3795
e-mail: varis@varis.com
web: http://www.varis.com

---

From: "Annette Murray", INTERNET:murray@dplus.net
Date: Tue, Jan 11, 2000, 9:59 PM

If you're interested try this experiment:

Create an 11x17 300 dpi RGB image.

Fill with Photoshop's Default Gradation: Blue, Red, Yelllow.

Dither On.

Convert to CMYK.

See the banding.

Try to eliminate the banding without adding noticeable noise.

Can you do it?

Create the same gradation but start out in CMYK.

Then Add Noise of 1 or 2 virtually eliminating all banding.

The gradation created initially in RGB is more vivid, but the banding is virtually impossible to eliminate in Photoshop.

This is an ongoing, frequent problem for us.

By the way, how do I convert my image to 16 bit, as recommended by several people? It is grayed out in my Photoshop Mode Menu.

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Tue, Jan 11, 2000, 8:45 PM

>By the way, is there any reason that the artist can't work in CMYK in the first place? This would take a lot of pressure off the back end of the process and minimize the need for handling most blends or grads with kid gloves.

>Kids gloves? Most darker gradients come out better if made originally in RGB. There are all kinds of problems using Photoshop filters on CMYK images.

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Wed, Jan 12, 2000, 12:32 AM

>Convert to CMYK.

>See the banding.

I only get banding when using the Built-In settings in CMYK Setup. When I use an ICC profile (I tried two, one for Matchprint, one for a printing press) I'm not getting banding.

---

From: "Annette Murray", INTERNET:murray@dplus.net
Date: Wed, Jan 12, 2000, 6:56 AM

> For contone blends we have a semi-solution.

> We make all blends (no matter what they'll be finally) at 0 - 100%. This gives the maximum number of steps that Photoshop will allocate.

> We noise that at 10% noise.

> Then we go to curves and adjust the blend to what it's densities should be.

> Try it, We think you'll like it.

Dear Lee,

I can create a beautiful bandless blend in Photoshop. What I am having trouble with are RGB blends, shadows, soft edges supplied by our customers (designers) that must then be converted to CMYK. At this stage horrendous, unfixable banding, posterization, contouring are introduced that noise alone will not eliminate.

Chris Murphy has suggested trying different ICC Profiles before making the conversion. He has had success. Hopefull the ICC Profile that makes the conversion without banding will not adverserly affect the color in the rest of the image.

---

From: Lee Blevins, INTERNET:leeb@ids.net
Date: Wed, Jan 12, 2000, 6:38 AM

Here's our experience with banding.

In vector blends, it's just rip dependent. Even at the same DPI, some rips do a better job than others. This must be relative to their internal math.

Our general feeling at Digital Graphics is that vector blends are what they are. We use both Rampage and Scitex rips and the result is about the same in both.

The big change for all of us on this is Postscript level 3. That put the vector blend problem to bed from what I hear. I don't personally have it yet but the change in internal rip math certainly makes sense as to what it being done.

For contone blends we have a semi-solution. We make all blends (no matter what they'll be finally) at 0 - 100%. This gives the maximum number of steps that Photoshop will allocate. We noise that at 10% noise. Then we go to curves and adjust the blend to what it's densities should be.

Try it, We think you'll like it.

---

From: INTERNET:shane@archangel.net, INTERNET:shane@archangel.net
Date: Wed, Jan 12, 2000, 6:12 AM

Hi Chris,

Thanks for pointing out that item about 16-bit precision processing for conversions. I had no idea that this was the case.

Since the results are so dodgy in gradients, I was prone to conclude that there might be even less than the usual 8-bits at work there.

---

From: INTERNET:shane@archangel.net, INTERNET:shane@archangel.net
Date: Wed, Jan 12, 2000, 6:07 AM

Chris Murphy wrote:

> Kids gloves?
Kid gloves. Just an expression, meaning gentle care and attention.
Comes from leathermaking origins. The softest leather is made from the skin of young goats. (Trivia)

> Most darker gradients come out better if made originally in RGB.
But if you stand the chance of getting steps in your gradient, and losing most of that colour anyway, is it worth it? I'm sure that's whole thread in itself!

> There are all kinds of problems using Photoshop filters on CMYK images.

Sure, some of them won't activate in CMYK space and others are unpredictable. However, there are lots of plug-ins that do work very well in either colour mode.

I work in a variety of spaces, depending upon the production path that my job will be following - and what I might need to do to an image to make it really zing. But, I always keep the final destination in mind so I won't end up fooling myself about how the final product is going to look.

Different strokes, I guess. (Just another expression :)

---

From: INTERNET:shane@archangel.net, INTERNET:shane@archangel.net
Date: Wed, Jan 12, 2000, 1:33 AM

Annette Murray wrote:

> By the way, how do I convert my image to 16 bit, as recommended by several people? It is grayed out in my Photoshop Mode Menu.

Hi Annette,

File must be flattened to do 16-bit.

On your suggested test, I only get very limited banding with 3 units of noise applied. The bands will show slightly on film or plate, but since they happen in high sat areas I don't think it will be visible on paper.

Obviously, Photoshop still lacks the type of gradient ingenuity that can be found in Scitex image editing products. Noise is an automatic feature of these systems. Granted, dithering of blends is a step in the right direction, but it will take a lot more engineering to produce the sort of consistency that designers are now demanding from this desktop product.

Making noise about noise might be a good idea.

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Wed, Jan 12, 2000, 1:29 AM

>How does this phenomenon you just picked up on relate to bit depth, if at all?

In this case it doesn't.

The conversion, even though the image is 8-bit, occurs through 16-bit precision; but that happens with the built-in option or the ICC option.

The bigger issue is that Adobe Photoshop separation tables are not well suited for separating this particular kind of image. More gamut compression is necessary in order to avoid banding/posterization at the sacrifice of reducing saturation of in-gamut colors. Photoshop separation tables were designed with compromise in mind, mostly that of getting the maximum amount of color (hue+saturation) retained from source to destination. The sacrifice is sometimes different colors in the source space (RGB) end up being mapped to the same or nearly identical values in the destination space (CMYK).

>What I don't understand is all this bit depth stuff as you mentioned in some previous postings. I understand what bits are and how they affect the amount of data, etc. But what difference does it make when we have to make it 8-bit in PS anyway? Why capture in such high bit if we have 8 or 16-bit limitations?

I'm not understanding the question. I think the original assertion about bit depth was that if you first convert to 16-bits then convert to CMYK, you'll find you're less apt to see banding. My point was that it has nothing to do with starting out in 16-bits before converting to CMYK because the conversion happens through 16-bits anyway.

The ONLY difference with starting out in 16-bits, converting to CMYK and ending up with 16-bit per channel CMYK image is that when you convert to 8-bits, Photoshop applies noise automatically during that conversion.

This does NOT happen when you start with 8-bits, the conversion through CMYK goes through 16-bit precision, and ends up as 8-bits. This additional noise has a "dithering" effect to prevent banding when going from 16-bits to 8-bits per channel and is a new feature in Photoshop 5.

>What is a good source for bit depth info?

Not sure of the author off hand, but there is a Real World Scanning & Halftones book that I'd guess talks about bit depth. Maybe not. Any other suggestions on books regarding bit depth?

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Wed, Jan 12, 2000, 10:41 AM

>Hopefull the ICC Profile that makes the conversion without banding will not adverserly affect the color in the rest of the image.

If it's an appropriate profile, if anything you will get a better separation than before. What will happen is a slight desaturation of colors that are in-gamut. ALL COLOR in the image is going to change (color = hue + saturation) in order to make room for out of gamut colors.

But of course if you select a profile that's not appropriate for the output conditions (just like if you use an non-ideal Photoshop separation table), you're going to get a non-ideal separation for the output conditions (and proofer), and end up with something other than what you wanted. ICC profiles are just a different way to make separations; they aren't intelligent and can't make intelligent decisions.

If you have an image that contains predominately in-gamut colors, but the gradient you are going to create will have quite a bit of out-of-gamut colors, then you can use two rendering intents. Create the gradient in a different document of the appropriate size, and convert that to CMYK while CMYK Setup is set with the correct profile and INTENT set to percpetual. Go to your image and convert to CMYK while CMYK Setup is set exactly the same except INTENT is set to relative colorimetric. Now you can bring the gradient into the original image.

So you get the benefit of gamut compression on only the area that requires it (the gradient), and you retain as much of a one to one color (little to no gamut compression) conversion from RGB to CMYK in your image.

It's false to assume that all of your RGB images contain large areas of out-of-gamut color with detail you care about. RGB as a whole is generally larger than CMYK in terms of tonal range and gamut, but this is misleading. What we care about is the IMAGE content. That is the better way to determine which rendering intent to use.

Using the perceptual rendering intent (gamut compression) on images that don't need it (i.e. they're already printable in the confines of the CMYK gamut) doesn't hurt the image, it just desaturates it very slightly trying to make room for out of gamut colors WHICH AREN'T IN the image. So if you don't have to do this, don't do it, use the relative colorimetric intent instead. You can think of this intent as what Photoshop separation tables use now.

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Wed, Jan 12, 2000, 3:18 PM

>Do you have some authority for this statement? I just did some tests with four gradients and four different methods of conversion--one Photoshop built-in, one Photoshop built in with a 16-bit file reconverted to 8bit afterwards, and two ICC profiles. The histograms of the images show thecharacteristic spiking and sawtoothing only in the three 8-bit conversions.

>From Chris Cox:

Built-in conversions go through 16-bit precision. ICC conversions go through 20-bit precision - REGARDLESS of whether the image is 8-bits or 16-bits (per channel).

Built-in conversions are ALWAYS relative colorimetric. ICC conversions are variable intent. That will also affect banding. Converting first to 16-bits, then converting to CMYK, then converting to 8-bits per channel uses the same precision as converting 8-bit RGB directly to CMYK except for adding 1/2 LSB of noise to the entire image.

So that's why it appears smoother. The precision is the same, it's just that noise has been added.

He assures me that the 8-bit and 16-bit conversions are identical because he copied and pasted 99% of the code.

---

From: INTERNET:shane@archangel.net, INTERNET:shane@archangel.net
Date: Wed, Jan 12, 2000, 3:08 PM

Just had a call from a friend who's looking for the best software to use for applying transfer curves to existing 16-bit CMYK TIFF files.

All suggestions welcome.

Thanks,

Shane

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Wed, Jan 12, 2000, 1:32 PM

>I can confirm that if I take an RGB gradation and then: Convert it to 16 bit; Convert to CMYK; and then Convert to 8 bit it is much smoother than making a direct RGB to CMYK conversion.

That's because the former method is automatically adding noise. I would expect it to be smoother.

>But I wish I knew what the difference between 8 bit and 16 bit is. And whyis 16 bit is offered as a Mode and what is it used for?

Bit depth is calculate as 2^x where x is the bith depth. 2^x is the way you write 2 to the X power (using superscripted x) with email since we can't make a superscript x above and to the right of the 2.

So 2^8 is 256. So 8-bits is the same as 256 levels. If you go to 16, it's 2^16 or 65,536 levels. Quite a big difference.

Now, when we say 8-bits vs 16-bits we mean PER CHANNEL. So for RGB, 8-bits per channel is referred to as 24-bit color (8 x 3 = 24). For CMYK, there are four channels so 8-bits per channel in CMYK is referred to as 32-bit color (8 x 4 = 32).

How many DIFFERENT combinations of RGB are there for 8-bit color? We know there are 256 levels per channel, and there are three channels. So 256 x 256 x 256 = 16,777,216 colors. You get the same number if you take 2^24.

This is the number of DEFINABLE points of color. You cannot device more POINTS than this for RGB. (For CMYK it would be 4,294,967,296 because it has four channels instead of three).

A lot of people get this confused and say, "well I can do 16.7 million colors, why can't I define the color I want? Humans can only see around 2 million different colors." The AMOUNT of definable colors has nothing to do with the GAMUT or the range of colors. If I want, I can define a color space of just red to yellow (no green or blue) and define it with however many millions or billions of points, but that doesn't mean those colors are different to the human eye - they are only different to the computer.

So 16-bits per channel gives us more levels, more definable points. For RGB that's 2.815E14, or 281,500,000,000,000 different definable POINTS.

To the computer these are different colors because each has a different value. So for gradients this is important.

Think of it like a staircase. If you have 2^8 colors per channel, 8-bit color, you have fewer steps available which means each step must be bigger. If you have 16-bit color, you have many more additional steps which means the distance between the steps can be much closer together.

The closer together the steps, the more "gradiated" your gradients.

---

From: Dan Margulis, INTERNET:76270.1033@compuserve.com
Date: Wed, Jan 12, 2000, 1:15 PM

Chris Murphy writes:

<<Did you create the RGB gradient in 16-bit mode? Or did you create the RGB gradient in 8-bit mode, THEN convert to 16-bit mode? If you created the gradient in 16-bit mode then of course it will be better. The question was if you have an 8-bit gradient, and you just convert to 16-bit mode before converting to CMYK, it will be better.>>

No, I created the RGB gradients in 8-bit.

<<I used the default built-in SWOP separation table in Photoshop 5.5 and got quite a bit of banding in a blue, red, yellow gradient (starting out in 8-bit mode, converting directly to CMYK). When I changed to a press profile I created, as well as a Matchprint profile I also created, I did not get any banding. There is far far far more variation among ICC profiles than there are among separation tables. I would generally expect with these kinds of highly saturated gradients to find Photoshop separation tables contributing to banding because they use very little gamut compression in comparison to an ICC profile.>>

Agreed.

---

From: INTERNET:amurray@ANROINC.COM, INTERNET:amurray@ANROINC.COM
Date: Wed, Jan 12, 2000, 1:10 PM

I can confirm that if I take an RGB gradation and then: Convert it to 16 bit; Convert to CMYK; and then Convert to 8 bit it is much smoother than making a direct RGB to CMYK conversion.

A can also confirm that different ICC Profiles produce different and varied banding in a conversion. Through many tests I am sure a good profile could be identified.

My preferred method is the 16 bit method.

But I wish I knew what the difference between 8 bit and 16 bit is. And why is 16 bit is offered as a Mode and what is it used for?

Thank you everyone!

---

From: Chris Murphy, INTERNET:lists@colorremedies.com
Date: Wed, Jan 12, 2000, 11:59 AM

>Do you have some authority for this statement?

Regarding 16-bit precision being used for these conversions, even if the image is 8-bit - as well as the noise introduced when converting from 16-bit mode to 8-bit mode - YES I have some authority. I got that from Chris Cox at Adobe.

>The histograms of the images show the characteristic spiking and sawtoothing only in the three 8-bit conversions.

>I see no significant difference in how pronounced the effect is in any of them. The 16-bit conversions are essentially smooth.

Did you create the RGB gradient in 16-bit mode? Or did you create the RGB gradient in 8-bit mode, THEN convert to 16-bit mode? If you created the gradient in 16-bit mode then of course it will be better. The question was if you have an 8-bit gradient, and you just convert to 16-bit mode before converting to CMYK, it will be better, and I find that difficult to believe since the conversion occur through 16-bit precision anyway.

However, I have forwarded your example to Chris Cox to see what he has to say about it.

>Also, that there is no significant difference in banding between using Photoshop built-in and an ICC profile, or at least not the two I tested.

I used the default built-in SWOP separation table in Photoshop 5.5 and got quite a bit of banding in a blue, red, yellow gradient (starting out in 8-bit mode, converting directly to CMYK). When I changed to a press profile I created, as well as a Matchprint profile I also created, I did not get any banding. There is far far far more variation among ICC profiles than there are among separation tables. I would generally expect with these kinds of highly saturated gradients to find Photoshop separation tables contributing to banding because they use very little gamut compression in comparison to an ICC profile.

---

From: Dan Margulis, INTERNET:76270.1033@compuserve.com
Date: Wed, Jan 12, 2000, 11:40 AM

Chris Murphy writes:

<<The ONLY difference with starting out in 16-bits, converting to CMYK and ending up with 16-bit per channel CMYK image is that when you convert to 8-bits, Photoshop applies noise automatically during that conversion. This does NOT happen when you start with 8-bits, the conversion through CMYK goes through 16-bit precision, and ends up as 8-bits.>>

Do you have some authority for this statement? I just did some tests with four gradients and four different methods of conversion--one Photoshop built-in, one Photoshop built in with a 16-bit file reconverted to 8bit afterwards, and two ICC profiles. The histograms of the images show the characteristic spiking and sawtoothing only in the three 8-bit conversions.

I see no significant difference in how pronounced the effect is in any of them. The 16-bit conversions are essentially smooth. They appear to be characteristic of having a curve applied to more data, NOT merely the addition of noise.

Accordingly, I conclude that converting to 16-bit first really does add precision, not just noise, to the process. Also, that there is no significant difference in banding between using Photoshop built-in and an ICC profile, or at least not the two I tested.

---

From: INTERNET:shane@archangel.net, INTERNET:shane@archangel.net
Date: Wed, Jan 12, 2000, 9:42 PM

There seems to be some disagreement as to when the task of eliminating banding should take place.

I'm in favour of introducing noise in the RGB file, prior to CMYK conversion. Some people feel that it's better to do it after you have a CMYK file.

Here's my logic...

Banding is the mapping of slightly dissimilar tones to the same tonal value within a given area wherein the (now identical) values touch each other to create a visible plateau of "sameness".

This is a common occurrence in CMYK conversion of gradients.

If we noise-up the given area and create a more statistically diversified area to begin with, the chances that NOTICEABLE banding will occur after CMYK conversion is dramatically minimized. This is because, due to statistical deviation, we have reduced the likelihood that now less-similar values within the given area will map to the same tonal value.

Very straightforward, I thought.

The contention on the other side of the argument is that banding will still occur after a noise-treated RGB grad is converted.

And it may. But a comparatively small amount of noise will kill it easily - especially compared to the amount of noise you would need to quell the effect entirely in the CMYK file. Less noise - same effect. Better image.

The human perception of "bands" in what should be a smooth transition of tone is predicated upon the adjacency of these tones to one another. Even though, on a microscopic level, the mechanics of what Chris and Chris are saying is true, at the level of human perception, the effect of noise-treating before conversion is more pleasing to the human eye. It's also more efficient, because you're working on a file that's 25% smaller in pixel data - and the work is done further upstream in the workflow. This almost always means a savings, especially if you're doing in-RIP seps from RGB data. In that case, you won't know you have a problem until you're looking at some plates.

My assertions can easily be validated in practical Photoshop examples on any home computer. Try it out.

---

From: INTERNET:shane@archangel.net, INTERNET:shane@archangel.net
Date: Wed, Jan 12, 2000, 8:05 PM

I wrote:

> <<However, if there were a conversion engine that applied statistically stochastic noise to the values being processed (mid-conversion), this might prove to be quite beneficial. <Nudge nudge wink wink to Adobe:>>>

Then Dan Margulis wrote:

> I'm not sure I understand what