Gradation banding remedies?

[Repairman]

Hi Folks
I'm comping an A1 poster, the background of which is a smooth gradation over the whole depth from white to 10% cyan. To avoid banding issues I normally add a bit of noise to disguise the effect but as white is involved I wondered if you have any better ideas? (8 BIT BTW).
R.

[Tony W]

R, somehow I think I may be missing the plot here and not understand. Are you saying that you already have the files for the comp. including the background?

Would it be possible to make a new background and use gradient fill foreground to background setting points from white to 10% Cyan. I tried this in sRGB 8 bit and on screen at least no sign of banding (well I could not find any!)

[0lBaldy]

Might need to convert to 16 Bit ..check this video..

[Repairman]

Depends a lot on the monitor Tony and it views differently on my three. However, we know the bands are there waiting to appear at the point of no return! (when money changes hands) It is not a worse case scenario but there are times when the blend is between dense and saturated colours, especially working in CMYK. I like the 16 bit solution (thanks for the link Baldy) but stayed clear in the past due to both the increase in file size and the loss of some tools. I guess if I create the grad in 16 and convert the file to add the 8bit elements the smoothness of the grad will be maintained?

[Chain]

Working in 16-bit will help you avoid banding in the first place (as it has more color values to pick from). You should not have visible banding on an normal 8-bit image if converted from a 16-bit image that had no visible banding (but do keep dithering on).

If you are not going to do any further adjustments to your gradient (e.g. curves adjustments and such), it should be fine to create it in 8-bit mode (dithering turned on).

If you have an 8-bit image that is already showing banding, possible solutions for fixing the issue would be to recreate the gradient and adding noise. Many techniques could be combined here, but I'm not sure what is the most practical way (could depend a lot on the image in question as well).

[Tony W]

R, I understand the concerns over the banding issue and I know that just a quick view on the monitor proves nothing you really need to be looking at what happens when the output device gets hold of it - I assume this is going to print.

I agree with Chain as my experience with computer generated 8 bit gradients has been fine providing you do not start making adjustments to that layer. Though I do tend hower to work in RGB with rare forays into CMYK!

The 16 bit solution however does sound like the safest route to take

[John Wheeler]

Tony, Chain, and OlBaldy have good advice about using 16 bit mode and looking at a sample from your final product to see if you can see banding.

Many monitors are only 8 bit per channel so even with a 16 bit image, it is effectively turned to 8 bit for viewing (where banding could be just artifacts vs real). It also turns out the if your ICC profile or graphics card is using a LUT (look up table) and calibration makes that LUT not a perfectly smooth transition curve), you can also see banding artifacts on your monitor yet potentially introduce real banding going to your printer. You can have the same issue with printers since many are just 8 per channel instead of higher. Not sure if you are going for an inkjet with RGB inputs or to prepress and separated CMYK channels. Going from RGB inputs to an inkjet that is using CMYK inks has another level of conversions going on in the firmware of the printer. Gotta go for a final sample from the printer that will be used.

That said, here are a couple more pieces/tidbits to consider.

- I do believe a small amount of colored Gaussian noise (0.1 to 0.2%) will be very helpful. Too much can become noticeable
- For A1 print size I am assuming that close examination at shorter than 10 inches will not be happening. If that is so, just make sure that the final print is at least 150 dpi. They eye will not be able to resolve pixels of that size at greater than about 10 inches so the eye will blend the noise as well
- Also there is a not well documented feature in Photoshop that applies dithering (1 bit noise) when converting from 16 bit mode to 8 bit mode (not the other way around). In Color settings with More Options expanded, there is a mode named Use Dither (8 bit/channel images) that adds 1 bit dither noise every time you convert between color spaces. It turns out that Photoshop adds dither under other circumstances if this option is checked (though that is not documented). One of those situations is when converting from 16 bit to 8 bit. If you have that feature on and convert from 16 to 8 bit (e.g. to save/print in JPEG) then you get a little dither noise automatically. This is like many other "auto" features in Photoshop. It may help yet is not under your control. I typically turn off that option (its on by default) and then add my own noise with a filter.

Personally, I bet you could get away with the background in 8 bit mode, adding some noise, and have enough initial pixels to print at a good dpi, and check your actual print given that actual printing is likely going to be in 8 bit mode anyways. Best practice would be to use the 16 bit mode as suggested by OlBaldy.

Let us know how it goes.

[Repairman]

Top quality replies as always from you particular contributors . On this occasion I am sticking with the noise option. CMYK dot rosettes will hide a lot sins and a smidge of noise will be my insurance policy. For more complex gradations I am going to follow the 16 bit route in future. John Wheeler how do you know so much stuff!
R.

[pixelzombie]

If the gradient is vector i will convert it to raster try and add noise but usually i have to recreate them in photoshop. I will usually add noise in the range of .5 to 1 pixel depending on the value of the gradient and then check it with a solar curve. I have also discovered that saving this file in any form of a jpeg will introduce its own form of banding so will usually save as a tiff or eps binary.

[mcdronkz]

This might be an interesting read.

[shift studio]

my favourite solution is the filter/brush strokes/spatter with radius 6-25 and smoothness = 1. It should break up any banding beautifully.
--shift studio

[Repairman]

Good link Mc and a neat trick Shift. Back in the day we used to output everything to 10x8 transparency - woe betide anyone who didn't pick up bands before output! The trick was to create a gradation using the same grey as start and end points; this would create a natural dither. Applying usm would 'pop' the grain after which you could make a contrasty B/W auto mask. LOL, retouching was more organic back then!

[chet webley]

Originally Posted by mcdronkz

This might be an interesting read.

That is very interesting!

[Chain]

Originally Posted by chet webley

That is very interesting!

Absolutely. Well written tutorial, and a very clever trick (It simulates Photoshop's 16-to-8-bit dithering "realtime" while working in 16 bit to prevent you from seeing banding on your monitor that isn't really in the image).

[Chain]

Originally Posted by shift studio

my favourite solution is the filter/brush strokes/spatter with radius 6-25 and smoothness = 1. It should break up any banding beautifully.
--shift studio

I just tested this on an image (best radius depends on the width of the bands). It worked great.

Attached example from an 8-bit image (a real example I had a while back). Banding was caused by a gaussian blur in a blue area. Then I applied Spatter. I ran auto-curves just to pump the contrast for visibility afterwards.

[shift studio]

Originally Posted by Chain

I just tested this on an image (best radius depends on the width of the bands). It worked great.

Attached example from an 8-bit image (a real example I had a while back). Banding was caused by a gaussian blur in a blue area. Then I applied Spatter. I ran auto-curves just to pump the contrast for visibility afterwards.

I use it in real-world situations all the time to smooth out large, close-to-flat areas - skys, walls, backgrounds - the trick is it must not be applied in areas where there is too much difference spatially, or you'll see ugly artifacts.

I also use it on masks that have banding in them

I'm attaching another image I posted when this topic came up in the past as an example.

--shift studio

[John Wheeler]

Thanks shift studio

I really like your technique and have added that to my tool box. It is really helpful for adding effective noise to the banding to have that spatial component added and not just noise on each pixel. Also thanks for the heads up about using it in other areas than the banding. I tried that out and it really messed up higher contrast edges. Your technique with adding a mask to control where it is applied is an excellent approach!! It's a keeper

[shift studio]

Originally Posted by John Wheeler

Thanks shift studio

I really like your technique and have added that to my tool box. It is really helpful for adding effective noise to the banding to have that spatial component added and not just noise on each pixel. Also thanks for the heads up about using it in other areas than the banding. I tried that out and it really messed up higher contrast edges. Your technique with adding a mask to control where it is applied is an excellent approach!! It's a keeper

I'm glad to have helped/contributed.
--shift studio

[Chain]

I just used the technique for an image that had bad banding issues (mostly due to JPEG compression).

I refined the technique by layering several layers with different radius settings. For the large area I made the noise smoother by using radius 23, 24 and 25, and blending them evenly. Close to the edges of a foreground object I used an incrementally smaller radius (masked).

The only annoyance I encountered was that the filter doesn't work in 16-bit mode.

[delene]

I have successfully used the spatter method for getting rid of banding on backgrounds in studio shoots but can anyone give me advice as to how to get a smooth transition from the subject (usually a baby crawling around!) as I always have difficulty in moving from the subject to the background.

Any advice on smooth transitions (generally using a mask) greatly appreciated!

[Chain]

If you were wondering how to use the spatter-technique close to edges/foreground objects, I would tell you to use a mask, and use a smaller radius the closer you get to the edge.

[delene]

Thank you. The artifacts are really bad though, and it will be a long job to complete, not really justified by the payment for the job. Thinking about the problem, I suppose my question should really have been "has anyone found a better way to get rid of banding on a plain background in a portrait shoot than the spatter method?"

[Chain]

Portrait or not, the spatter method works well in areas with smooth transitions (that has some banding in it).
If you have harsh artefacts and banding I would recommend applying a gaussian blur first to smooth it out. This might solve the problem for you (especially if you are working in 16-bit mode).

If it's too damaged, you might want to recreate it using a gradient or a large soft brush.