How Do I Work With 8 Bit Images Without Posterization

[randy_myers3]

<p>I've done some searching here and just haven't found a good answer. I'm afraid that's probably because there isn't a good answer but here goes anyway.<br>

Yes, I'm aware that one "should" work in 16 bit and I ALWAYS do for my own stuff but for clients, I'm very frequently limited to 8 bit images because that is all they have to supply. Most civilians just don't know about this stuff.<br>

For instance, I'm doing the graphics for a CD cover right now. Luckily the client knew enough to send me images that were large enough to work with but all of the different source images were in 8 bit. So, I'm now on the home run stretch as far as finishing this job but I have an image that could REALLY use some punching up in the way of curves or levels adjustments. I'm finding that even with what I would consider to be very modest adjustments, the resultant image ends up with a histogram with a whole lot of pretty major holes the entire height of the histogram.<br>

Now, I see absolutely no posterization on the screen but I've read that even in that case, it could very well show up in the print.<br>

So, here I sit after doing about 10 hours of work on this project and can't really finish up the last couple minutes of work necessary for good contrast and color without screwing up the histogram.<br>

I don't want their prints to possibly be posterized but I also don't want to send them a fairly flat looking image.<br>

Any ideas? I sure hope so because this happens on a semi-regular basis and it would be GREAT to know what to do when it rears it's ugly head again.</p>

[richardsperry]

Convert to 16 for editing?

 

Copy the image and paste it into a 16 bit new document?

[tom_mann1]

<p>If I'm going to do any substantial amount of PP on a JPG, and there is no way I can get my hands on the corresponding RAW file or a 16bpc TIF of the image, I bring the JPG in through ACR set for 16bpc output, only performing tasks in ACR like perspective adjustment which actually help fill in the gaps between 8 bpc levels. Once in PS (at 16 bpc), I'll usually give the image a dose of PRP's "Histogram Repair" ( http://powerretouche.com/Histogram_Repair_plugin_tutorial.htm ). </p>

<p>If you don't have PRP's Histogram Repair, you can just add a tiny amount of random noise or fake grain. Such approaches clearly do not work in all situations, but they certainly help break up banding in less drastic PP situations.</p>

<p>I'll keep the image at 16 bpc in PS as long as humanly possible, only converting back to an 8 bpc JPG as the very last step (if a JPG is needed).</p>

<p>HTH,</p>

<p>Tom M</p>

[randy_myers3]

<p>Richard,<br>

Thanks. I gave that a shot before posting this. Converted to 16 bit. Did a minimal amount of editing but when I converted back to 8 bit, I still ended up with the holey histogram.</p>

<p>Tom,<br>

Thank you. I will check out your link and noise techniques to see if that helps. <br>

Also, please forgive me for my run-on question. I forgot about whatever it is that I have to do in order to keep my paragraphs and spaces intact on this site?</p>

[patricklavoie]

<p>theres no need to work in 16bits all the way.. whats important is that all the color and processing of the file (density, contrast, saturation etc) been done in your raw developper in 16bit pro photo.. then when you export, if you dont have any MAJOR color and luminosity issue to fix, working with a 8bits file is perfect (i do it all the time) you can work in Adobe RGB or even in sRGB at that point if your final destination is commercial printing like in your case (the file will look good also when print on your epson anyway)..</p>

<p>Take note that the histogram repair trick is more for a client that have read that on the web than anything else.. i mean, it wont do much on your image itself.. it only fill the gap in your histogram.. not a big deal seriously.</p>

<p>If you have banding on your image when working, i would suspect my raw development, how far you push the contrast or mainly the saturation of your color.. and when transform to 8 bits, that can show.. but honestly, if you work correctly withtin limits that you will gain with experience, working with a 8 or 16bit wont change much the end result.</p>

<p>Even if in mathematic 16bits is far superior, and working in Pro Photo is THE way to go.. i rarely see the need of it most of the time, and many retoucher i know, and even some well knowed one dont work in 16bit and many dont work at all in Pro Photo (after the raw processing i mean). Im retouching high end image for beauty, fashion, cosmetic.. far more difficult than your usual band image or mom and dad one.. and for those, i definatly dont see any use for 16bits.</p>

<p>*as Tom mention, if working with JPEG (major issue with quality and artifact depending of who shoot it, who developped it, is it from a under exposed file etc..) bringing this 8 bits to 16bit could help staying away from banding .. good explanation from Tom already . ; )</p>

<p> </p>

[lachaine]

<p>What's converting the 8-bits per channel file to 16-bits per channel going to do?</p>

<p>I think it's a moot point anyway, because the gaps in the histogram are meaningless... and even more so if you are resizing that image to be smaller than it started out as. Just don't overdo the curves, levels or whatever.</p>

[patricklavoie]

<p><em>What's converting the 8-bits per channel file to 16-bits per channel going to do?</em></p>

<p>it will in many circumstance save you from any banding issue *visually* when working with jpeg (sky gradient in particular, other than that .. nothing really special *visually*) .. and sometime you wont see any difference at all on most images that dont have a strong gradient ; )</p>

[randy_myers3]

<p>Patrick,<br>

Thank you very much for your help. I understand that working in the raw developer would be ideal and is what I would do with my images but in the circumstance that I am describing, I'm give 8 bit images from the client so I have to work with them from the beginning.</p>

<p>Pierre,<br>

So, when I do the levels changes that I would like to do on this image, I go from the first histogram to the second. Are you saying that I should not worry about the second histogram and that it probably won't cause banding in the final print?<br>

<img src="http://img62.imageshack.us/img62/8719/1histograms.jpg" alt="" /></p>

[patricklavoie]

<p>do you see banding on your monitor? if not, i really doubt they will be any in the end.. from what i see here, a simple *histogram repair* will cover that.. for nothing probably, but will cover it.</p>

<p>go to filter / noise / add noise / enter 1 / monochromatic / gaussian.. bam. done.</p>

[randy_myers3]

<p>Thank you very much. I'll stop worrying about it then.</p>

<p>I've never had anyone say that they had posterization/stair stepping on a print but I made the mistake of reading somewhere that when you see a histogram like the one above, even if you don't see the problem on the screen, it still might show up on the print. I've been worried about it ever since. I guess I'll try to stop thinking about it. : -)</p>

[dave404]

<p>If you pick up the Dan Margulis books on Photoshop he only uses 8-bit and laughs at people that expect perfect histograms. It only matters when the gaps and spikes get really wide, otherwise nobody is going to notice. That said I shoot RAW and stay 16-bit as long as possible. It give you more room to work with.</p>

[randy_myers3]

<p>Thanks David!</p>

[digitaldog]

<blockquote>

<p>If you pick up the Dan Margulis books on Photoshop he only uses 8-bit and laughs at people that expect perfect histograms. </p>

</blockquote>

<p>He’s so wrong. His testing methodology in this context is super weak too (see http://www.brucelindbloom.com/index.html?DanMargulis.html).</p>

<p>Lets see, we’ve had high bit scanners for decades as well as support for this in Photoshop and other imaging applications as long. We have high bit capture in our camera systems. Adobe, Nikon, Canon, the big high end scanners are all wrong, Dan’s right about 8-bit per color. Please! The math is undeniable too. Can you see the effects? Depends. For Dan, ink on paper low linescreen like his books, no. On a device that has the quality (like my 10 color Epson’s)? Sure can. </p>

[randy_myers3]

<p>Hi Andrew,<br>

Sorry, I didn't see your post until today.</p>

<p>Wow, this argument goes back a LONG ways I see. For the life of me I can't imagine anyone asserting that making color corrections in 8 bit would be equal to or even BETTER than doing them in 16 bit. That goes against every single thing that I have read on the subject in the last 6 years.</p>

<p>Unfortunately, I'm stuck having to do corrections on the 8 bit images that I am supplied. It seems fairly obvious what that does to the histogram. My problem is that I don't do any printing myself so I don't know ahead of time whether or not a histogram with blank spaces with render banding or stepping in the final print. It's VERY rare that I see any of these problems on the monitor so that leaves me guessing.</p>

[tom_mann1]

<p>Hi Randy - Sorry it took me a while to get back to this thread, but here's another way of saying what some of the other participants in this thread have been saying about the relationship between banding and histograms. </p>

<p>A histogram is what is called a "summary statistic", ie, it summarizes the statistics of the entire image. If there is banding in one area of the image, a histogram of just one channel for just that one area will likely be very spiky. However, the luminosity histogram (ie, a composite of all three color channels) for the entire image will likely be much smoother because the RGB values of pixels in other parts of the image might easily fill in the gaps and pull down the spikes from the problematic areas. </p>

<p><strong>In other words, an image with obvious banding might have a relatively smooth histogram.</strong></p>

<p><strong>Conversely, even if the histogram of an image shows huge spikes and gaps, there still may be no important visible effects on the image. </strong></p>

<p>Posterization is likely to be visually unimportant when the image doesn't have any large, smooth areas. As an example, attached is a discarded image I took a few nights ago on a family outing. I intentionally posterized it to about 20 levels and show the histograms for both before and after posterization. I see no visually important differences between the posterized and non-posterized images. </p>

<p>Histograms are incredibly useful, but when it comes to issues such as banding / posterization, IMHO, about the best that can be said is that the presence of spikes and gaps in a histogram definitely tells you that you are losing quality, but this loss of quality may or may not be visually important. </p>

<p>HTH,</p>

<p>Tom M</p><div></div>

[tom_mann1]

<p>And here is the same junk image, but this time, I intentionally posterized it to about 20 levels. To my eye, the differences between the two versions are almost invisible.</p>

<p>However, if there had been large featureless areas in the image, the posterization would likely be very easy to see.</p>

<p>Tom M</p>

<p>PS - Unfortunately, it appears that downsizing and saving for web smoothed out much of the bumpiness of the histogram of this image, but the conclusions are the same -- ie, I couldn't see any differences when comparing the two versions in PS.</p><div></div>

[randy_myers3]

<p>Tom,<br>

Thanks for doing this work in order to demonstrate what you're talking about.</p>

<p>The thing that bothers me is that I've been told that even though an image may look just fine on the monitor, that doesn't mean that the posterization won't show up in the print.</p>

<p>I guess when there is a REALLY bad histogram, one never really knows if there will be posterization until the print is made. Unfortunately, I don't have a printer and would hate to have a print made every time that a histogram looks a little scary.</p>

<p>I guess I'll just take my chances and wait until a client says, "HEY, I HAD A PRINT MADE AND IT LOOKS HORRIBLE. WHAT HAPPENED?" : -)</p>

[tom_mann1]

<p>Hi Randy - I haven't run into a case where the print was made at a reputable printer (eg, MPix, not a kiosk at CVS) and there was banding on the print that wasn't visible on my monitor. I guess it could happen, but I haven't seen it. OTOH, I recalibrate my monitor every couple of weeks and periodically check the results of the calibration by looking at images I'm familiar with, test patterns that check specifically for banding, etc. I think that having a monitor that you can trust is a huge factor in this. </p>

<p>The way I think about histograms and banding is that if the histogram looks really bad, it's a "red flag" and I should proceed cautiously, not do any extensive processing on the image, etc. If there's a really bad histogram AND there are important smooth areas in the image (eg, skin, sky, etc.) it's a another red flag and I proceed even more cautiously. If I see banding or posterization problems on my monitor, I rework the image before I even waste time on a test print.</p>

<p>Also, unless it's a real emergency, I never send a print to a client until I've seen it first. The one time that, because of a time crunch, I recently had to have some prints shipped directly from the printer to the client involved a set of big posters - a couple of thousand $$$ job. I have to say that my heart was in my throat until I finally saw them for myself. OTOH, I have worked with this printer for years and he told me they looked OK before he sent them on their way.</p>

<p>Cheers,</p>

<p>Tom M</p>

[randy_myers3]

<p>Thank you very much Tom.</p>