Using an LCD screen as a 'digital negative' in alt-process contact prints

[AndrewBurns]

I spent a bit of time making 'quality of life' improvements to my setup. It's all the same parts but now they're all mounted to the light module and run from a single power supply so I can move it around more easily. Eventually it will be mounted to a board with some holes so that I can use pins to register the screen and paper for multiple exposures.



I also got a large piece of ground glass to try my light-field correction again and the results this time were much better! This is my first attempt, and there's obviously still a vignette around the edge to clear up, but the vast majority of the non-uniformity between LED's is gone.



Original on the left, corrected on the right.

Hopefully now when I print images with a lot of large middle-gray areas there won't be a noticeable pattern...

 

[koraks]

That looks quite acceptable by any means, I'd say. Even the vignette cleared up for the most part.

 

[AndrewBurns]

For completeness, here's an actual photo with the light-field correction applied. This is a photograph of the image the LCD and UV light source is projecting onto ground glass (obviously not inverted like it would be for printing).



To my eye there's no unevenness in the large areas of continuous tone, which means that the correction is working (ignore the light stripe along the top edge, that's there in the original photo). This photo looked really bad when printed previously because of the uneven light field, so hopefully this time with the correction it will print well. I've just coated a few sheets so hopefully this afternoon I can get back into printing instead of fiddling with the equipment!

 

[koraks]

That's a major improvement for sure. This probably solves it for 99.9% of the images.

 

[AndrewBurns]

Unfortunately life is never so easy, the results are definitely better but when I print the photo above there are still a few really obvious squares of non-uniformity and it's driving me completely insane. I've spent all day trying to figure it out, I suspect that the LEDs change intensity as they heat up over the course of an exposure in a non-uniform way which could change the correction but I haven't been able prove if that's an issue or not.

There seems to be something fundamentally different about taking a photo of the image projected onto ground glass vs. printing it, because the variation in the print is large enough that I would think it must be more visible in the ground glass photos. The cyanotype process is obviously very non-linear, meaning that very small differences in exposure create large differences in tone on the print in the highlights, so the light source needs to be extremely even, but even so the ground glass photos would seem to indicate it should print fine. I'm slightly suspicious that the intense UV is confusing my camera somehow...

Ultimately this is very frustrating as the really interesting part of this system is supposed to be the LCD screen which so far works perfectly, the light source was never supposed to be the problem! If I bang my head against this wall for too much longer I'm going to have to think about changing the light setup completely but I'm not sure what would work better at this stage.

 

[koraks]

Well, I remain of the opinion that the best solution is to increase the distance between the light source and the prints. It'll make the setup less compact, but the problem with uneven illumination will go away.

 

[AndrewBurns]

The difficulty is that the LEDs are still quite well collimated even after the diffuser, as evidenced by how sharp and square the non-uniform patches on the prints still are, increasing the distance would make them less obvious but it would take quite a lot of distance to make them blended enough to not be noticeable (so much so that it might be practically impossible without stronger diffusion).

Very early testing showed that if the light was too diffuse the print lost a lot of sharpness, but I wonder if what I did back then was really a fair test, as I don't think I put much effort into making sure the print was held flat against the screen. I could try just removing the lens assembly entirely so the light is very diffuse and see how that goes, but my suspicion is there's enough thickness of glass between the part of the LCD that masks the light and the print that I'll always loose too much sharpness following that approach, like putting the negative the wrong way around in a contact print or not having a good printing frame.

The people who make these units include the collimating lens for a reason, but I suppose in the 3D printers they're designed for the thing that you want to expose (bath of resin) isn't perfectly touching the screen and so the collimation may be more necessary than in my application.

 

[koraks]

Try 20 to 30cm between the print and the LED source and remove the collimator plate / lens array. At that distance, everything will blend together quite well with bare LEDs and collimation will still be good enough to image the grain on silver halide film. This is basically the setup I've been using for a year or two now. I've tested various LED arrays and this combination of bare LEDs and a little less than a foot of distance seems to work consistently well.

 

[AndrewBurns]

Yes if I really can't get the current setup working I'll have to give removal of the collimating lens a shot, that will definitely get me even light, it will just come down to what it does to sharpness.

That said I think I've made a bit of a breakthrough, check out these two photos:



Do you see the distinct squares on the right-hand image that aren't there in the left? Those are what are currently showing up on my print even after correction. Both of those photos are the same uncorrected light field, the difference being the first is the camera photographing the light source directly and the second is through a pair of cheap laser safety glasses that came with my laser cutter.

The squares are not at all visible when I look at the screen wearing a pair of UV-blocking laser safety glasses, but they're clearly visible with the unknown pair, which I suspect mostly block IR and visible wavelengths but pass some UV.

My theory is that the camera will almost certainly include an IR and UV cut filter inside, so I've mostly been correcting the visible component of the light source but not the UV that actually impacts the print. The cheapo laser glasses are mostly blocking visible light, so the exposure time is very long (30 seconds or so) but most of what makes it through is actually the UV which is also what effects the print. Hopefully if I make my correction based on this it will actually improve things.

 

[AndrewBurns]

After a substantial amount of banging my head against the wall I had to take some time off from this project to recover. Ultimately the correction process using the camera didn't work just because the camera (even with random plastic filters) doesn't respond at all accurately to the wavelengths of light that alt-process prints do.

I decided to have another crack at it this weekend and have made some good progress now. I decided that the only way to accurately record the wavelengths the print actually cares about was to make prints. So I made a print of a flat grey rectangle, which very clearly showed variation in density relative to the variation in the light source. I then took a photo of this print with a copy stand and used it as the flat-field correction image (after converting to B&W, flipping, inverting and applying my cyanotype calibration curve to it). I made another print of the same flat-grey rectangle but with the correction applied and the vast majority of the non-uniformity is gone.

Uncorrected on the left, corrected on the right.


Obviously not perfect yet, there's a lot of negative-vignetting which is probably a result of how I digitised the image (digital camera on a copy stand with two LED light panels either side) and some weirdly dark areas top and bottom. I'm quite confident I can fix these issues however, I'll just repeat the process of digitising the 'corrected' print and add that to the first correction image so that the result is a combination of both.

I knew that this was probably the best way forward early on but I was really trying to avoid it as it's quite slow, having to wait 24 hours for the print to dry and darken each time I want to iterate, but glad that it's working out.

 

[koraks]

Well, that's a creative solution for sure, although it doesn't sound particularly robust to me. Individual aging of emitters might necessitate re-calibration later on.

I do realize that the use of the LCD and the inherent distance between the 'digital negative' and the print surface necessitates better collimated light than a traditional contact frame that can keep the print and the negative in direct contact, with no LCD cover etc. between the two.

I think I'd be inclined to solve the issue differently and more along the lines of Calvin Grier's approach to exposure: with a powerful light source several feet away from the print frame (horizontal projection). It's seen in this video of his:

He uses a big vacuum frame and some sensors for light integration, but the concept is straightforward. The actual light source is an array of COB LEDs.

 

[AndrewBurns]

Yes as you say there are quite a few ways that it could 'drift' away from calibration over time. I also wonder if I would need to re-calibrate for different processes that might have different sensitivity vs. wavelength curves (e.g. I'm calibrating for cyanotype currently, but would DAS for carbon printing respond the same or need re-calibrating?). My hope is that aging effects over time will largely effect all LEDs equally and any non-uniform brightness changes will be small vs. the magnitude of correction I'm having to apply initially, but I suppose time will tell.

I've seen videos of Calvin doing his exposure with that setup and it certainly is interesting, seems like a fairly wasteful way of doing things but simple and robust. I'm not sure I'd want to stand in the same room as my light source for a full exposure time (currently 4 minutes) if it was uncontained. The LCD screen only transmits about 10% of the light at most, but when I've tested the light without the LCD in place even for a fraction of a second you can feel a significant amount of heat 'bathing' your skin. I have very light skin and a family history of skin cancer so I'd probably get sunburn in short order.

 

[koraks]

seems like a fairly wasteful way of doing things but simple and robust

Yes, well, 'wasteful' is relative if you keep in mind why he does it this way. For is gum and also carbon prints, he needed very good collimation to prevent the effect he calls "blowdown", which is the collapse of tiny little dots of pigment-loaded colloid (10-20um diameter in his case, I think). It's a bit of a rabbit hole to go into why and how that's an issue, but the net result of all of this is that he needs very good collimation and the combination of some focusing using crude lenses and large distance is practically the only really feasible way to accomplish this. I imagine you could get a similar effect with a grid (much like you'd use on a strobe), especially a fairly deep one, but this, too, would be very wasteful.

you can feel a significant amount of heat 'bathing' your skin

For wavelengths we use, in the range of 365-395nm, the risk to your eyes is more relevant than skin. You'd have to spend a LOT of time directly in front of your exposure unit to accumulate even a fraction of the skin irradiation you'd get when taking a brief walk on an unclouded day. Still, it's good to be careful. Wear sunglasses with open systems like Calvin's, and don't linger unnecessarily directly in front of the unit.

 

[AndrewBurns]

Well the process seems to be working. This print has just been hung up straight out of the developing bath so it will lose a lot of that texture and gain some contrast as it dries down.



I might need to see if the light field correction I'm applying is also adding extra texture to the print, I blurred the photos I used to generate the corrections but I might need to blur them a bit more if some paper texture is carrying through into the correction image (which would then get overlaid onto the image I actually want to print).

 

[AndrewBurns]

I mentioned in the first post I'd tried directly exposing via moving a laser point around and obviously I'm currently exposing using a wide area light through an LCD screen but there's also a third option which is also used for 3D printing and industrial applications like photo-lithography, which is digital light projection (DLP). Basically bouncing a strong UV light source off a special microchip which is actually an array of millions of little moving mirrors. Each mirror can be controlled to either reflect the light source down through a lens or off to the side onto a blank surface. The end result is you can project a binary image, and by rapidly moving the mirrors back and forth you can increase the bit-depth of the image to make a greyscale.

The reason I didn't go down this path is that it's much more expensive and the resolution is currently limited to about 4k at best while my current LCD is 7k. The low resolution can be overcome by tiling multiple small images together, which I'll have to do with my LCD for prints larger than A4 anyway. The big advantages are efficiency (because the light loses very little energy being reflected vs. about 90% loss going through the LCD screen) and you're just using a single high-power LED rather than an array so the light source is going to be inherently much more even and consistent.

I'd be interested to see how low of a resolution you can get away with in an alt-process print anyway, I'm currently at about 550 dpi and I've not been able to see any pixels on my prints with the naked eye even looking very closely. Inkjet prints tend to be around 300 dpi, so I wonder if even something like that would be perfectly adequate, particularly in a multi-layer process where slight variations in overlapping registered layers would probably mask the appearance of any pixels even at lower resolutions.

Not really an update, just food for thought.

 

[calebarchie]

Andrew, there was a light engine released not that long ago that may be suitable and cheap for what it is. I think it was 8K I will buy it and try eventually - don't think resolution requirements need to be that high.

Typical inkjet prints are 300 or 600ppi but the half-tone dither to achieve that ranges from 2400-5000dpi (that is the nozzles on head itself).

 

[AndrewBurns]

Well I think I've finally got this process/system dialled in and I'm making some (what I think are) pretty good prints! Here's a dodgy phone photo of two I have hanging up to dry now, I intentionally exposed an unfinished edge after the main image exposure to help add back some 'organic' feel to the otherwise pretty clinical prints.



Soon it will be time to go bigger by tiling together multiple prints with a pin registration board...

 

[AndrewBurns]

Well it's been a while. I finally got around to attempting to 'tile' multiple exposures from my LCD screen setup into a single larger print. Obviously I'm limited by the size of the LCD screen I have (about A4 or 8x10 print size), so to do anything larger I have to get creative.

I made a registration board with pins that locate the LCD exposure unit relative to the paper, and I can pick up the exposure unit and put it down on different pins that let me 'tile' exposures together. I've designed the registration board to provide some level of overlap between the tiles, with the idea of blending the tiles together so that any small misalignments are less visible.

Here's my first attempt at an A3-sized test pattern:


Clearly there are a few issues:

• The underexposed line down the middle is where the exposures are supposed to overlap and add together, I forgot to account for the significant non-linearity between LCD screen density and print density, meaning just linearly blending the images isn't sufficient. I'll have to use my calibration curve in the blending.
• The two exposures don't line up very well, I must have made a mistake somewhere in translating the physical overlap into number of pixels
• There's a dark line down the middle of the print, the LCD screen is leaking UV light out one of the edges which I'll need to cover up
• If you look closely I printed the mouse cursor of my computer on both exposures!

Still, it's a start, and I think with a little work I'll be able to get it a lot better. The registration board will also allow for the tiling of 3 exposures rather than two for printing on A2 paper.

 

[koraks]

If you look closely I printed the mouse cursor of my computer on both exposures!

Haha, that's neat! I had to look pretty hard for it though!

This looks very promising; not bad for a first try.

 

[AndrewBurns]

Having a think about it for 2 seconds, my blending issue is a simple matter of order of operations. I know that there's a strongly non-linear relationship between LCD screen density and print density, and I apply a calibration curve to the original image to compensate for this, but I applied my linear blending gradient after the calibration curve. All I need to do is apply the blending gradient before the calibration curve and then it should work fine (as long as I get my tile overlap and alignment correct as well).

 

[AndrewBurns]

I made another test print of a pattern I designed to check for alignment of the overlapping tiles. This is an A3 sized print with two tiles that overlap about 11mm vertically. If the overlap was perfect the central line should be solid, but as you can see the tiles misalign by about 10 pixels (~0.5mm) which is honestly better than I would have expected. Better still, vertical translation and rotation look close enough that I don't think I need to compensate for them, so once I fix the overlap amount it should be good to go.

 

[AndrewBurns]

Well I haven't really been able to get the overlap blending working, I'm definitely missing something about how it physically works. As you can see in the following test prints, the left and middle prints have an overlap down the middle blended together but the overlap area is too light in dark tones and too dark in light tones (and also not even across patches of the same tone). The print on the right is with no attempt at blending and a small overlap.



I've since attempted to cut the blending region between tiles way back to only 4 pixels and rely instead on very accurate alignment between the tiles, and the results have been much better. This print didn't quite get the tiles close enough together so there's a gap of less than 10 pixels between them.



This one I moved the tiles 4 pixels closer but there's still a small gap, another 4 pixels might do it. There's also slightly darker stripes down the print where the LCD physically overlaps, I think there's an issue in my code where the parts of the screen that should be totally black are still transmitting a little UV, so these areas are getting a tiny bit more exposure than they should (this should hopefully be an easy fix though).



Overall I think I'm pretty close to getting acceptable results. Frustrated that I couldn't figure out the tile blending issue, but my registration setup is repeatable enough that aligning the tiles within a pixel or so (~50 microns) should be achievable.

 

[AndrewBurns]

After the issues I was having with my linear-gradient overlap I thought of a different type of overlap I've decided to call 'stochastic blending'.

The issue with the linear gradient blend between tiles is that it relied on a perfectly linearised printing process, because you're basically printing two different tones that have to add together to make a third tone perfectly. If the process isn't linear then the two tones you print don't equal what you actually wanted and then the tone you get when you add them both together is also wrong in unpredictable ways.

So rather than adding together the overlapping pixels I thought why not just mesh them together in a way that any given pixel in the overlap region is either on or off. That way you're not relying on having a perfectly linear process.

To do this I've made an image mask where for each row of pixels in the overlap region, a number of them are randomly turned on or off. In each subsequent row of the mask more pixels are randomly turned on or off, so you end up with a gradient of active pixels, from all of them at the start of the overlap to none of them at the end. And the other image tile is simply the inverse of that, so when they're both joined together the full image should be displayed with no gaps.

I don't think this technique will be as resilient to misalignment as a linear blend, but I'm hoping the pixels are small enough and the light source has enough diffusion that a little misalignment won't be noticeable when spread over a large enough area.

 

[AndrewBurns]

Well I'm still having issues and also coming to the realisation that it's never going to work unless I can get some overlap printing correctly. I've been getting within a few pixels of having the two tiles butt up against each other perfectly but the registration is never perfect enough to prevent a 1 or 2 pixel gap or overlap which is clearly visible (eyes are great at seeing straight lines where they shouldn't be).

Also the 'stochastic blending' worked perfectly in software but not so well in the real world. I experienced a similar issue when trying to print halftone images with my screen, because the light source isn't perfectly collimated and the screen 'mask' is relatively far off the paper individual pixels either blur away to nothing or block together with neighbouring pixels.

There is in effect a threshold of how small of a point can actually be printed, the screen will happily only allow a single pixel worth of light through, but the light that hits the paper will diffuse to a large enough area and low enough intensity that no pigment actually gets generated at the paper. So when trying to blend images together on a per-pixel level the isolated pixels never actually get printed and you end up with a light stripe down the middle of the page.

The only way I can think to get around this limitation at this point is to make my blending coarser, so that I only break the overlapping regions into clumps of 4 or 9 pixels which will hopefully print well but still allow for some blending of minor misalignments without it being visible to the naked eye. Ultimately the whole 'tile multiple exposures together' thing could just prove too difficult though, maybe with a lot more effort I'll figure it out, or maybe it's just not feasible in which case I'm limited to printing ~8x10s.

 

[koraks]

Frankly, and not to be discouraging, but at this point I think I'd throw in the towel. To be honest, I had my doubts about this tiling approach from the start. Overlaying a couple of layers on top of each other with film & paper (or film & film) is one thing. But doing the same thing with an exposure unit...let me put it this way, there are companies in the semiconductor industry making billions upon billions of revenue on the basis of this competence.