Fixing Really Bad Moire Patterns

<p>Anyone have some good tips on fixing bad moire patterns? Not the blue/yellow coloration, because I've researched several good ways of getting rid of those, but the patterns that are left when you get the coloration out. Below is a sample.</p>

<p>I've played with overlapping channels, but the pattern is usually in all the channels. If not, anyone have good ideas for patching/cloning clothes texture?</p><div></div>

<p>Is this a scan or from a digital camera?</p>

<p>If it is from a scan, you need to use ANG glass, start wet mounting or use a holder that does not touch the scanner glass.</p>

<p>If it is from digital, you need to use a different camera. Most cameras have a filter on the sensor to illiminate this. However some manufacturers leave it out, on the belief that sharper and higher resolution will result. Leica, I believe, is one of the manufacturers that leave it out.</p>

<p>Cloning...maybe. If this is a common problem see Peter's response above. You really can't "fix" it.</p>

<p>I don't know that this is a "good" solution, but I have had luck doing the following:</p>

<p>Create a new layer and set the blending mode to "color." Select a good color from the coat and paint it over the color in the moire. To address the pattern issue, I've had luck dodging and burning (new layer, set to overlay, fill with 50% gray, paint with low opacity black and white or use the dodge and burn tools) the pattern. Cloning is going to take some real patience, but it can be done. I would zoom in very close, clone at 100% opacity, and build it in slowly.</p>

<p>Also, I'm not familiar with how sensors differ from camera to camera (at least not in regards to moire), but moire is simply the result of the pattern in the clothing matching the patterns in the sensor (in camera), or the pattern in the clothing matching the pattern on a display device like a monitor. In the latter case, the moire is only visible at particular zoom levels and will not show in print. In the former case, the pattern is as much a part of the image as the person is. A slight zoom in or out, or slightly changing your shooting perspective then eliminates the problem. If you have multiple frames, check to see if it is in all of them. Unless you shot with a trip and a fixed lens, you may have snagged an image without it. It doesn't take much change in a single variable for the pattern to disappear.</p>

<p>Good luck, and let us know how you handled it and how the results came out. I don't believe you are going to be able to find a quick fix. I've always tackled it with patience and time.</p>

<p>Nadine,</p>

<p>You can get rid of the blue/yellow variations simply by desaturating (since it's a gray suit). The luminance variations will be a bear, though!</p>

<p>What sort of camera took this shot? The AI filter's spatial frequency rolloff is clearly too high.</p>

<p>This is from my 5D (original), not a scan. I know the background behind in camera moire and I know that one cannot really do anything about it when shooting, except use another camera. Changing distance, angle etc.,--yes, it helps--but one can't totally eliminate the occurence without changing the camera (sensor).</p>

<p>The above is from a shoot I did a couple weeks ago. I didn't get the moire in all the shots, and in some, it isn't as bad. The above is the worst instance. I was kicking myself because I've had this happen once before, but with a suit that was comprised of small checks. I didn't think this suit would cause a problem, and didn't zoom in enough to see it.</p>

<p>The moire is definitely in the file--it isn't just monitor or screen appearance. I know I can get rid of the yellow/blue/rainbow coloration, but the dark pattern is the problem. Looks like I'll be cloning, etc., I also wondered about taking a patch of material, pasting it in, and warping it. Jen--I did the dodge thing with my previous escapade with moire, and it works to an extent, but it is very tiresome and time consuming. Probably going to have to do it again.</p>

<p>Anyway, thanks for the info so far. Anyone else have any good tips about removing the moire <strong>pattern</strong>, I'm all ears.</p>

<p>First, do the partial desat fix, as described above. </p>

<p>Next, work on the the brightness / luminosity data. Instead of cccloning, select the bad areas and feather the edge of the selection. </p>

<p>Then, go crazy inside the selected areas with either the patch tool or one of the healing tools. Do this on a separate layer at 100% opacity. When done, drop the opacity and lots of the moire in luminosity will disappear. Fixing the problem with the patch or a healing tool can be done a lot more roughly and quickly than with the clone stamp tool.</p>

<p>Tom M.</p>

<p>I think you can reduce or eliminate the problem with a different RAW conversion (browse...I forget which). In other words, the problem may not be so evident in the original file. I'm surprised to see your results from a camera so famous for massive internal processing. <br>

I've found sharpening of Pentax files (which are almost as unprocessed as Leica M8) in Lightroom leads to moire (not as bad as your example), but that doing the sharpening in PS2 after B&W conversion via Lightroom seems to solve the moire problem.</p>

<p>...in other words, you may want to explore other RAW converters. </p>

<p>Nadine, just an off-the-wall thought: I haven't seen a 100% crop to see exactly what's going on, but you MIGHT be able to kill some of the moire pattern and keep the texture by using a "salt and pepper" filter. At least that's what the filter is called in PaintShop Pro. You can set the size of the speck to be filtered and set certain threshold parameters for detection. You'd undoubtedly have to set some rather strange values for those parameters. It's a long shot, but who knows?</p>

<p>John--thanks, I will try something different--I processed the above in Lightroom. Will try DPP and Camera Raw.</p>

<p>Sarah--thanks. I did something similar to get to the stage above. It is: apply strong gaussian blur, add noise, change blending mode of duplicate layer to color. Is that different from what you're saying?</p>

<p>color moire is easy to get rid off;</p>

<p>1_double the background</p>

<p>2_apply a strong gaussian blur to it</p>

<p>3_change the blending mode of this layer to color</p>

<p>4_put a mask on it and remove with a brush where you want the effect of the color moire remove.. work everytime on anything any color.</p>

<p>BUT your problem is pattern moire.. way harder to get rid off. I suggest, oh god.. i have a hard time saying it.. use C1 Pro.. arghh just saying it its killing me!..but for that kind of problem, C1 Pro is your best bet.. forget ACR it will not give better result than Lightroom...</p>

<p>Im also surprise that you have such a problem with that camera, im use to see that on a medium format camera due to abd light / distance and other.. but on a canon 5D.. first time ( i own 2 of those )</p>

<p>Patrick--why C1 Pro? I know they have a moire plug in but I heard it works, but not completely. Or are you talking about the RAW conversion process itself?</p>

<p>I'm not worried about the color, but the pattern that still remains after the color is gone. I just tried the channel part of the following. You use the healing brush but use the green channel (the least moire showing) as a pattern. It actually works pretty well. In the above, the green channel wasn't completely free of dark stripes, but the result is a lot better than not using it, and I can use the dodging tool to lighten the remaining stripes up.</p>

<p><a href="http://www.shootsmarter.com/index.php?option=com_content&task=view&id=68&acat=15">http://www.shootsmarter.com/index.php?option=com_content&task=view&id=68&acat=15</a></p>

<p>Nadine, I was thinking of something quite different. If you post a 100% crop of the moire pattern here, I'll play with it just a bit to see if my long-shot idea might work. Others might also have a chance to work out some manipulation.</p>

<p>I, too, am quite surprised your 5D did this. I'd really be interested in seeing the 100% crop to learn, at a microscopic level, exactly what happened. I'd also be interested in any close-up you have of the fabric pattern. I wonder if the only way to prevent this might have been to use a diffusion filter or a really soft lens. Weird. Definitely something to learn from!</p>

<p>C1Pro to develop the shot only.. then run to Lightroom for the rest of your life ; )</p>

<p>Sarah--I'll get a 100% crop ready to post. Thanks.</p>

<p>Patrick, but why C1Pro to develop? What does it do differently?</p>

<p>OK--here are two 100 percent crops.</p><div></div>

<p>I did what I suggested in my earlier post to your RH 100% crop. The only change I made from the procedure I outlined in that post was that after I did a bunch of applications of the patch tool, too much of the high frequency info (ie, texture) was suppressed, so I had to put in a couple more layers to recover the HF texture.</p>

<p>To recover the texture, I made a copy of the image immediately after the color moire had been removed. I then ran it through a high pass filter ( r=0.7 px), spread out the histogram of that layer by moving the endpoints in, changed the blending mode to "overlay", duplicated the layer, and placed both copies above the layer created by the patch tool. Using such a small radius in the HPF ensures that the lower spatial frequency luminosity variations caused by the Moire pattern are not picked up by the HPF, only the fabric texture.</p>

<p>I duplicated the high pass layer because doing only one overlay of it was not strong enough to fully recover the texture of the cloth. Two layers were a bit too much, so I dropped the opacity of the 2nd HPF layer a bit. I then applied the original area selection mask to these layers so that only the area with the Moire that had been patched was re-texturized. The result is shown below. If I had spent a bit more time on this mask, the transition between the fixed and unfixed areas would have been smoother (eg, as the Moire pattern comes up to the lapel).</p>

<p>The tonality and sharpness in the fixed area is still a bit off. Going back and re-adjusting opacities and curves would have improved this, but since this was just a demo of the concept, I didn't think this was necessary.</p>

<p>The bad news is that this approach is labor intensive. It took me about 12 min to bring the image to this level. The next time through the process, this approach would go faster, but it's still always going to be labor-intensive.</p>

<p>HTH,</p>

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

<p>It can't be removed, the only thing that could be done is painting over it with a clone brush.<br>

As others said, try a different raw converter. It should be possible to eliminate it in raw conversion, however the result may be a bit softer. However, I don't know if there is any raw converter that currently exists that will do it :-)<br>

During shooting, you can try to use a different aperture or lens to get rid of it.<br>

This is actually very good as an example, since many people say that AA filters on cameras are not necessary and that moire can be removed in post. Well, it can't be removed in post, except by painting over the areas, but that's a lot of work.</p>

<p>Nadine, my original long-shot idea was a no-go. I was expecting a different sort of moire pattern (and was prepared to be unbelievably impressed with the sharpness of your lens). I had always wondered about the possibility of a color moire pattern such as this one and had just never seen it happen. The only way to prevent this sort of moire pattern would be to use a much more conservative rolloff frequency in the AI filter, where the nyquist rate is based not on the density of photosites, but rather on the density of the bayer groupings. Oddly, I can also almost see a justification for a 4-color system, as the G channel would not be doubly represented. This problem would also be completely avoided with a foveon sensor camera. Fascinating.</p>

<p>Anyway, my thoughts were that the luminance info is basically intact between R, G, and B photosites, but that there may be overall response differences between the three, thus mucking up a simple desaturation approach. Thus if one were to tweak the output of the three channels independently, the luminance might then even out across the hue shift areas. The easiest way to do this is with the channel mixer. I ended up using 24% R, 33% G, and 43% B, but I wasn't completely happy with the results. The problem was that I had no idea what really went into the demosaicing algorithms and therefore couldn't reverse it. Anyway, you can try that mix if you want to see what it does. It's better than using the G channel luminance, but still not quite there.</p>

<p>I got much closer to the mark with a differing strategy. I thought I might be able to use the G channel if I could only shift the hues to make the hue variations symmetrical around green. Instead of a yellow/blue, I wanted more of an orange/cyan. I rotated hues by -12 and THEN broke out the green channel. It looked quite a bit better:</p>

<p> </p><div></div>

<p>The problem I ran into was reflected by the fact the R channel seems to end up with a higher contrast than the B channel. I suspect this has something to do with the demosaicing algorithms and/or the relative selectivity of the response curves (R, G, and B) with regard to the spectrum from the fabric. Maybe more tweaking needs to be done in the degree of hue shift. The balance in your original photo actually yields a hue shift between roughly 20 and 150, centered roughly over 85, which would be green. So my theory here might not be entirely spot-on. However, the approach still shows promise.</p>

<p>I would think a next step in experimentation would be to adjust the contrast of the red channel to more closely match that of the G and B channels, then to hue shift, and then to break out the green channel. Once you get the magic formula down, you should be able to crank out all your photos.</p>

<p>Obviously, then you'll need to create a separate layer for the suit to superimpose over the rest of the frame.</p>

<p>This is quite a nightmare for a pro photographer! Please tell me this: Can you see the moire pattern in the LDC playback at full magnification? In hindsight, how could you have recognized this problem before it became a PROBLEM? This is certainly one of those situations where it would have been nice to have film. (I say that as a 99% digital photographer!)</p>

<p>because C1Pro have the best method of removing camera moire.. maybe youres is to extreme, but Phase One really have a great software for this... and that come from a Lightroom aficionado that will never go back to C1Pro.. but some case still require it ; )</p>

<p>Sarah, I'm impressed. Very nice work with good theory behind it!</p>

<p>Tom M.</p>

<p>Wow, Sarah--I'm going to have to read your posts several times before I begin to understand your thinking process, not being a scientific or mathematical type, in fact, quite the opposite. Thank you so much. I will try to see what I can do.</p>

<p>Your questions: yes, I can see the moire pattern on the LCD but only when magnified, and it does not appear as strongly--less contrast and lighter. Since I had an incident like this once before, I failed at recognizing the problem again :^)</p>

<p>Last night, I brought out several pieces of clothing with small patterns to try to duplicate the response, thinking that if I could identify something that caused the moire with my 5D, I could test my 40D to see if it would also produce moire. Since the sensor is different, I reasoned (this basic reasoning is all my poor brain can do) that if I could even be aware of the possibility, I would just switch to my 40D. Both cases involved young men's suits. I photograph bar and bat mitzvahs. I'll see if I can upload a sample of the previous one so you can see. I would expect the size of the pattern to be a major similarity. I couldn't find a piece of clothing that produced the moire, and since I don't have much access to boy's suit material, I will need to hunt around...</p>

<p>Patrick, thanks. And Tom, also thanks. Looks nicer that what I can do so far. I know I have a lot of time ahead of me, trying to find a solution.</p>

<p>Nadine, Sarah's approach approach blows the doors off of mine both in terms of quality and amount of labor. Use her approach.</p>

<p>Instead of doing a hue rotation before breaking out the green channel, it is almost equivalent to use one of the color-selective BW conversion tools and tweaking the colors to minimize the banding in the luminosity. I tried both the BW conversion tool in the NIK Color Efx Pro package, and the native BW conversion layer in PS. The NIK product was considerably more effective in completely reducing the luminosity bands, but the PS native tool was quite acceptable and is available to anyone with PS, so I showed this in my example below.</p>

<p>The nice thing about Sarah's approach (or the above variants) is the huge saving in time compared to even my patching / high-pass filter approach (which itself is much faster than a painting or cloning approach). For example, I can fix one of your example images with just 3 adjustment layers, as illustrated below.</p>

<p>Sarah: again - a wonderful idea! Congrats!</p>

<p>Tom M.</p>

<p> </p>