Need help getting rid of canvas texture

[mancub]

Hi there,

I'm new to the forum and a bit of an amateur to PS. I mostly use PS for editing photos and am currently running CS2. I had a scan done of a painting of mine and the light is reflecting off of all the canvas threads, is there a way that I can eliminate this textured effect? I did a little searching and found the FFT filter, but the end result for me was no different. Below I've attached pictures of the who piece, a close-up detail showing the canvas threads, and a studio shot of how the piece looks normally.

I would appreciate any and all advice with this problem.

Thanks!
Jacub

 

[Tom Mann]

By far, the best way to get rid of the pattern is to photograph it correctly in the first place. No post-processing approach will do as well as this.

So, assuming that you don't have the original artwork any more, so you can't re-photograph it using proper technique, then, of the various post-processing approaches, the FFT approach is, by far, the best.

I'm not sure why your attempt didn't work, but presumably there was either something wrong with your install of the filter or your operation of it. Each of the implementations of FFT plugins is slightly different, but here's a good tutorial on the Joofa version:
https://www.youtube.com/watch?v=SUqsm9_iCbQ

Unfortunately, I don't have time at the moment to process your image this way, but, if you don't have success, perhaps I'll be able to get to it late tonight (USA east coast).

Cheers,

Tom M

 

[ALB68]

"By far, the best way to get rid of the pattern is to photograph it correctly in the first place. No post-processing approach will do as well as this."

What am I missing here? The OP had it scanned and not photographed. Would photographing it give a better result than the scan?

 

[mancub]

By far, the best way to get rid of the pattern is to photograph it correctly in the first place. No post-processing approach will do as well as this.

So, assuming that you don't have the original artwork any more, so you can't re-photograph it using proper technique, then, of the various post-processing approaches, the FFT approach is, by far, the best.

I'm not sure why your attempt didn't work, but presumably there was either something wrong with your install of the filter or your operation of it. Each of the implementations of FFT plugins is slightly different, but here's a good tutorial on the Joofa version:


Unfortunately, I don't have time at the moment to process your image this way, but, if you don't have success, perhaps I'll be able to get to it late tonight (USA east coast).

Cheers,

Tom M

Hi Tom,

Thank you for the reply. Depending on where I'm sending my art I have a couple places that I use to either professionally scan or photograph it using high resolution, large format machines. I recently moved cities and due to a tight deadline, had to use a new place to "professionally scan" my work. I was less than impressed with the results, but the paintings needed to be shipped in order for them to make it to the show on time. With the show coming up in week I have to send the gallery some images to use for the preview. I'm hoping after the show to send them to my regular photographer, but for now these are all I have to work with.

I will watch the video you linked and give the FFT approach another attempt. One thing that I noticed that was different than the other examples I saw online was that during the process I never had a large central star. There were some smaller stars that I blacked out scattered in patterns, but they were very hard to discern from the grainy background, and only a few of them. I will attempt it again and take a screen shot to show you what I mean.

Thank you again for your reply, it's very much appreciated!

Best,
Jacub

 

[Tom Mann]

That's right -- artwork that has surface texture (eg, impasto techniques, textiles), or which exhibits shiny, specular reflections from the top layer of a smooth painted / glazed surface, or even an impasto surface is almost always re-photographed (not scanned) by museums for their catalogs, archives, for reproduction in art books, etc.

Even putting aside issues of size, the reason is that one has vastly more flexibility when photographing such pieces compared to scanning them. The most common example of this is to use polarizing films over your lights, and then an optical quality polarizing filter (rotated at 90 degrees to the ones on the lights) in front of the lens to reduce specular reflections and to bring out the colors. This is done essentially for the same reason that one uses a polarizer when photographing bodies of water with the sun reflecting in it. In addition, one can adjust the lights to be far to either side of the camera so that their reflections (ie, off of flat painted surfaces) do not hit the camera, but bounce away, harmlessly. For textiles (including canvas), one adjusts the angles of the lights to be near grazing incidence if you want to bring out the texture, or closer to the camera and using large diffuse sources of light (eg, softboxes, umbrellas, etc.) if you want to minimize texture as the OP does.

HTH,

Tom M

 

[ALB68]

That's right -- artwork that has surface texture (eg, impasto techniques, textiles), or which exhibits shiny, specular reflections from the top layer of a smooth painted / glazed surface, or even an impasto surface is almost always re-photographed (not scanned) by museums for their catalogs, archives, for reproduction in art books, etc.

Even putting aside issues of size, the reason is that one has vastly more flexibility when photographing such pieces compared to scanning them. The most common example of this is to use polarizing films over your lights, and then an optical quality polarizing filter (rotated at 90 degrees to the ones on the lights) in front of the lens to reduce specular reflections and to bring out the colors. This is done essentially for the same reason that one uses a polarizer when photographing bodies of water with the sun reflecting in it. In addition, one can adjust the lights to be far to either side of the camera so that their reflections (ie, off of flat painted surfaces) do not hit the camera, but bounce away, harmlessly. For textiles (including canvas), one adjusts the angles of the lights to be near grazing incidence if you want to bring out the texture, or closer to the camera and using large diffuse sources of light (eg, softboxes, umbrellas, etc.) if you want to minimize texture as the OP does.

HTH,

Tom M

Thanks for the explanation.

 

[Tom Mann]

...One thing that I noticed that was different than the other examples I saw online was that during the process I never had a large central star. There were some smaller stars that I blacked out scattered in patterns, but they were very hard to discern from the grainy background, and only a few of them. I will attempt it again and take a screen shot to show you what I mean....

Hi Jacub - The absence of the central star is bizarre. It is the so-called "DC component" of the FFT. It carries the average luminance information. It's absence would indicate an average luminance of zero! However, perhaps this is part of the particular FFT implementation that you used, and the user (or a call to some other function) is supposed to add back in the average gray level, after this step. I would consult the supplier of the FFT plugin about that.

The weakness of the peripheral stars is troubling, but less so than the absence of the central star. These localized Fourier components (ie, the peripheral stars) are often weak simply because the resolution of the image is a bit low and the texture is somewhat blurry. This does not at all appear to be the case on the sample you posted, but perhaps something else is going on. Also, have you checked the maximum pixel dimensions your plugin can support. With some FFT implementations, if your image is too big, the data will "wrap around" in the memory buffer and produce gibberish.

Please let me know how it goes.

Best regards,

Tom M

PS - BTW, which FFT plugin are you using? Windows or Mac? 32bit or 64 bit?

 

[Tom Mann]

FWIW, I dug out my ancient freeware FFT (32 bit) plugin and installed it on my 32 bit version of PS CC for the PC. I then applied it to your image. I saw the expected pattern of dots, blacked them out, applied the inverse transform, and came up with this:



It isn't great -- there's still quite a bit of smearing, but probably not as bad as one might get with traditional techniques for noise removal. Also, if I had spent a bit more time on it, and had the full image to work on, I suspect the results would have been even better. Anyway, just wanted to demo that the FFT technique does indeed work on your image.

It's up to you whether you use it of not. Since people looking at the photo obviously realize this is a reproduction, perhaps they will accept a bit of the weave texture in the photo, so it might be good to apply the reduction at full opacity to the featureless parts of your image, but at lowered opacity for the areas where there is detail.

Tom M

PS -FYI, the name of the plugin that I used was FFT-3d4x from 2008.

PPS - If I had thought ahead, I would have added a gray border to your image before I ran it through the FFT software. This would have put the ringing artifacts seen inside the boundaries of this image out into the added gray zones, which then could have been cropped away.

PPS - ALB68 - Re use of photography for fine art reproductions, check out note #1, all the way at the bottom of this tutorial: http://retouchpro.com/tutorials/?m=show&id=185 . There certainly are situations where one can do just fine with a scanned version, but other cases where it would not be the best choice. You need to have an experienced person make the call on this.

 

[mancub]

FWIW, I dug out my ancient freeware FFT (32 bit) plugin and installed it on my 32 bit version of PS CC for the PC. I then applied it to your image. I saw the expected pattern of dots, blacked them out, applied the inverse transform, and came up with this:

View attachment 50850

It isn't great -- there's still quite a bit of smearing, but probably not as bad as one might get with traditional techniques for noise removal. Also, if I had spent a bit more time on it, and had the full image to work on, I suspect the results would have been even better. Anyway, just wanted to demo that the FFT technique does indeed work on your image.

It's up to you whether you use it of not. Since people looking at the photo obviously realize this is a reproduction, perhaps they will accept a bit of the weave texture in the photo, so it might be good to apply the reduction at full opacity to the featureless parts of your image, but at lowered opacity for the areas where there is detail.

Tom M

PS -FYI, the name of the plugin that I used was FFT-3d4x from 2008.

PPS - If I had thought ahead, I would have added a gray border to your image before I ran it through the FFT software. This would have put the ringing artifacts seen inside the boundaries of this image out into the added gray zones, which then could have been cropped away.

PPS - - Re use of photography for fine art reproductions, check out note #1, all the way at the bottom of this tutorial: . There certainly are situations where one can do just fine with a scanned version, but other cases where it would not be the best choice. You need to have an experienced person make the call on this.

Tom,

Wow, that looks much better, thank you for taking the time to test the filter on the detail that I shared. I know now that it is probably something that I did. You mentioned that if the image was too large that the data will "wrap around", this is possibly the case as I was working off a 35mb Tiff. I will reduce the size of the image and start over again with your suggestions to see if that clears things up. I used the Windows 32 bit plugin.

I'll post again after my next attempt. Cheers!!

Jacub

 

[mancub]

Please let me know how it goes.

Tom,

I reduced the image size to 1200 pixles across and right away could see the center star and orbiting stars when I applied the FFT filter. I followed the video that you sent a link for, although my plugin doesn't say Joofa. I followed along right till the end where he was deleting the layers that weren't green, but I wasn't able to do exactly what he did. My end result was a 'ghost-like' image of the painting. Am I on the right path or did I mess up entirely? I know you've already spent a lot of time replying to my questions, so if I'm asking too much of your time just say so . Thanks again for all your help, your assistance has been much appreciated!

Best,
Jacub
www(dot)jacubgagnon(dot)com

 

[mancub]

UPDATE:

I just wanted to post a big THANK YOU to Tom Mann - this guy really saved my Canadian bacon! I stumbled across the forum last week looking for advice editing some pretty terrible scans I had of my art, and Tom came to the rescue! I worry about sharing high res. images of my work online, it's a scary world for visual artists in this day and age, but Tom was very professional and right from the beginning I found it easy to trust both him and his abilities. With Tom's help and knowledge of FFT filters (Fast Fourier Transform), he was able to take the canvas texture out of my images. I was then able to make a few slight adjustments, and voila, my crappy scans never looked better!

I still can't believe the help and generosity that I was met with upon finding this forum, I'll definitely be following along on PSG from now on and I hope to one day return the favor in kind.

Thanks again, Tom for taking the time (around the holidays, no less) to help a stranger in a tight bind!

Cheers!!!!

Jacub

 

[Tom Mann]

Thank you very much for the kind words, Jacub. I'm glad to hear that I was able to help out, especially considering the art show that you are participating in down in Florida.

For anyone following this thread, minimizing the canvas texture was a very interesting exercise for me. I haven't needed to use the FFT for a year or two, and my past uses of it was for de-screening half-tone line screens when trying to salvage irreplaceable photos that were now only available from scans of newspaper clippings.

Canvas is a bit different problem. The most important difference is that the canvas texture is nowhere near as regularly spaced as a half-tone screen. This means that each of the "dots" in the Fourier plane for a canvas scan is much larger and has wings (in the directions of the axes) much broader than in the case for offset printing.

Secondly, the canvas texture "rides over" all areas in the image, everywhere from the dark, featureless areas to areas of normal brightness where there is important texture that absolutely must be preserved. What this forces one to do is carefully consider which spatial frequencies (ie, bright crosses in the 2D FFT) need to be attenuated, and by how much. So, the key "trick" in processing Jacub's canvases was not to zero out very low spatial frequencies -- only the higher ones, and then, only very small frequency ranges around each "dot".

The rest of the process is very similar to how one uses FFTs to de-screen offset press printed photos, and, in case anyone is interested, there are quite a few tutorials available on the web about de-screening using FFTs.

Another trick is that to reduce edge "ringing" artifacts, first double the size of your canvas, then perform the FFT, apply the filtering (ie, darkening the dots), do the inverse FFT, and then just crop away the edges. This old trick works quite well, but I felt we needed even better results, so I used what is called "periodic boundary conditions". Basically, I added areas to the photo in exactly the same way one would do this to make a "canvas wrap", except in this case, the added areas were huge -- they quadrupled the area the FFT had to handle. The benefit to doing this is that because of the perfect symmetry on either side of each of the 4 edges, the FFT process introduces essentially no edge ringing artifacts.

Attached is a very small section of one of the two images I worked on. It shows that while the reduction in canvas texture is substantial, it is intentionally far from perfect. I let some texture remain to reduce smearing (as compared to the first image I posted in this thread).

Again, best of luck, Jacub, and please do come back to this forum (for instruction, or to teach us some of your tricks), or to our freelance forum (for paid work). However, I suspect that because you are already reasonably skilled in PS, you'll probably be less interested in the latter.

Cheers,

Tom M

PS - And just to reiterate my original comment in this thread: "if you can rephotograph the artwork using good technique, you do vastly better than trying to fix a problem after the fact".

 

[ALB68]

I have to say..this is very heady stuff. My compliments to Tom for his knowledge in this area. As learned as our membership is, I doubt many here (including myself) have an understanding of how this works. Good job Tom !