WIP Repair Tutorial WIP Guide on restoring backglass art, with notes & examples

Isaac Sauvage

Meep!
Site Supporters
Joined
Jun 23, 2003
Messages
6,815
Solutions
10
Reaction score
2,877
Points
150
Favorite Pinball Machine
Jurassic Park
Below are my thoughts and working methods. I'm using GIMP, but Photoshop and many programs work the same way.

How to reconstruct a skewed photo?
1) Crop the image as closely as possible without losing any of the artwork. Note: If a backglass is sitting in the backbox, some of the art is likely going to be cut off by either a frame or the edges of the wooden backbox. I usually try to crop right to the edges of the visible art instead of accidentally including any frame edges. Because redrawing the lost edge areas and reconstituting the original size is a lot more work, usually not necessary.
2) Use GIMP's Transform / Perspective tool to pull the visual corners of the artwork all the way to the physical corners. Start with the most skewed corners to help better adjust the other corners. Zoom way in to make sure of your accuracy. Circle around the four corners of the image at least 2-3x to make sure everything is adjusted just the right amount. Save, and you're done. Note that this approach will automatically correct a certain amount of 'fisheye.'
|
Note: If the image has lots of fisheye, and you don't want to lose a bunch of edge
area, you'll want to start step one without fully cropping. Instead, turn on grid
mode and try to crop about 3-4 units larger than you normally would. The extra space
created here will allow you to preserve the bulging edge sections for the time being.
Perform step two as usual, aiming to unskew the image evenly, again leaving a four
pixel margin around the corners. And then:
|​
3) Go to Filter / Distort / Lens Distortion and adjust either the "Main" or "Edge" slider until the edges are straight. A gentle hand is required here, as you only want to correct the edges, not throw off the rest of the image or alter the corners which you already corrected.
4) Finally, you can crop away the margin to produce your finished piece.
5) If necessary, you can then resize the image according to accurate backglass dimensions using references like this. Just remember that even with the straightening work you've performed, the image you produced is probably missing some of its margin if it was photographed in the machine's backbox. So you might want to correct for that, depending.

How to take a small source image (with issues) and use AI to expand it properly?
BigJpg.com and Waifu are fantastic for this, as exampled below.

How to auto-remove obvious artifacts?
Enhance filters starting with Noise Reduction are the usual way, but the two tools mentioned just above are super-easy to use and incredibly effective. Almost like magic, really.

Note that the more an image is composed of distinct lines and uniform colors, the better the two sites will work on these issues. Also note that it's best to use just enough noise reduction to clean up. Using too much will lose detail in your image.

How to remove yellowing or specific color oversaturation?
Choose Colors / Hue-Saturation, then select the color you want to work on. (note: if your color isn't directly based on the six RGB / CYM colors, then you might need a different method) Drag the Saturation bar to the left to decrease that color. Then drag the Overlap slider to the right in order to make a smoother transition from your base color to similar colors, for example green and red if you're working with yellow.

How to restore contiguous areas using surrounding areas as a source?
There are in fact lots of methods. To maintain a certain level of color & texture variance, the Clone tool followed by the Smudge tool is a classic. If a single fill color works, then bucket fills set to a sufficient transition can work. The paintbrush with various brush settings can also work, such as the airbrush and/or using non-100% brush opacity. Maybe easiest of all is to simply select, copy, and paste from a nearby region, then smooth as necessary. Heal tools can also work, particularly for more complex areas, by auto-using surrounding regions to reconstruct the focus area.

When is it worth re-creating by vector tools?
Probably not necessary in most cases, as it can be a lot of work, plus it's hard to capture the color & detail nuance of the original in a really satisfying way. That said, older and simpler backglasses will work the best, and can probably be 'textured' to look more realistic as the final step.

Reconstructing playfields by vector tools is probably time better spent, as they tend to be simpler than their corresponding backglasses.

What's a fairly easy, effective way of reducing glare & light reflections?
I don't quite understand the methods I'm seeing so far, likely because I haven't used them before. Would be really nice if there was a filter or smart tool I could apply to an area, but I'm not sure it's that simple.

How to remove watermarks?
So far the "sharpen by synthesis" tool I installed has been disappointing. Main issue with PBA's stuff is that their watermarks work by decreasing gamma / saturation in order to create lettering. Problem is-- so far the anti-watermark stuff I've Googled assumes that the lettering is dark or black, and therefore that the way to fix it is to pull in surrounding elements to reconstruct the 'destroyed' letter area. *Bzzp* Note: I also want to point out here that the object is not to steal PBA's stuff, potentially cutting in to their business model, but simply to help improve backglasses to be used in pin-sim projects.

Hmm, anything else...?
 
Last edited:
Maybe this online AI image enlarging site might be of use?

Not much of a graphics man myself, though I have used the clone tool and yeah does the job but can be a right pain in the wrist.
Funny, I was just about to use that site, but wanted to see if I could remove JPG artifacts first. Because what's the point of enlarging artifacts, y'know?

So I played around with Gaussian Blur a little bit in GIMP and felt like I was getting nowhere.

Googled a little bit more and found THIS amazing tool. Check out the *&^@# job it did on this backglass loaded with artifacts!! (used the "medium" setting)

Before and after:
(you'll probably need to zoom in on the first one to see all the issues)
 

Attachments

  • ccm_football2.jpg
    ccm_football2.jpg
    83.5 KB · Views: 43
  • ccm_football2_waifu2x_art_noise1_tta_1.png
    ccm_football2_waifu2x_art_noise1_tta_1.png
    376.4 KB · Views: 50
Last edited:
Nice find Nic, bookmarks site...anime graphics tools, who would have thought?

Those two sites combined could improve a lot of images and tables.
 
Nice find Nic, bookmarks site...anime graphics tools, who would have thought?

Those two sites combined could improve a lot of images and tables.
Okay, I just did a comparison test between those two sites. Enlarged x2 in both, with no noise reduction options. "BigJPG" took ~2 minutes and produced a 1.2mb result. "Waifu2" took about 10secs and produced a 1.1mb result.

Both results look amazing, and even better than the earlier version with cleaned artifacts. They've managed to turn jagged edges and such in to clean, crisp lines. (I wonder what the upward size limit is?)

So I'll have to test more, but maybe I can get away with using a single pass on Waifu2 to both reduce noise and enlarge x2. Yes, these tools are absolutely incredible. Altho note: evidently they work less well on photos and highly-detailed pieces.

Once again, here's the original image and the cleaned, double-sized final image:
 

Attachments

  • Football (Chicago Coin, 1949) (BGResto).jpg
    Football (Chicago Coin, 1949) (BGResto).jpg
    83.5 KB · Views: 39
  • Football (Chicago Coin, 1949) [960x1056] (Waifu2).png
    Football (Chicago Coin, 1949) [960x1056] (Waifu2).png
    1.1 MB · Views: 40
Still amazed how the right steps can take the attached image all the way to the linked one.

-crop (GIMP)
-unskew (GIMP)
-reduce noise (Waifu/BigJpg)
-enlarge (Waifu/BigJpg)
-sharpen (GIMP)


I know there's more stuff (and I mean processes, not hand-editing) I could be doing, like playing with levels, channels... uh, stuff like that. For example, in terms of sharpening, I don't yet understand this section very well:

To prevent color distortion while sharpening, Decompose your image to HSV and work only on Value. Then Compose the image to HSV. Go to Colors → Components → Decompose.... Make sure the Decompose to Layers box is checked. Choose HSV and click OK. You will get a new grey-level image with three layers, one for Hue, one for Saturation, and one for Value. (Close the original image so you won't get confused). Select the Value layer and apply your sharpening to it. When you are done, with that same layer selected, reverse the process. Go to Colors → Components → Compose.... Again choose HSV and click OK. You will get back your original image except that it will have been sharpened in the Value component.
 

Attachments

  • BallyMalibuBeachBINGOBG.JPG
    BallyMalibuBeachBINGOBG.JPG
    489.7 KB · Views: 35
Last edited:
You got me interested in this. I tried "Waifu2" but it has a significant file size limitation. "BigJPG" can handle larger file sizes. I tried it on the playfield of The Walking Dead. that I consider problematic. I reworked it and imported it into the FP table and made screenshots. below that are cropped The results are better than I expected although like you said the parts of the playfield that use photos aren't improved as much.

This is the original:
Walking Dead, The ProX 1.08.png

This is the same place on the playfield using "BigJPG" set to the highest noise reduction and 2x. I tried medium and high artifact removal and preferred the highest setting.
Walking Dead, The ProX 1.084.png

Then I tried the highest noise reduction at 4x, perhaps slightly better.

Walking Dead, The ProX 1.08.png
 
Nice work, George.

- Theoretically for an image like that, one could mask out the illustrated sections from the photo sections and run them through separately, then recombine in GIMP / PS as two layers in to one. The point of all that would be to use BigJpg's different "Artwork" / "Photo" settings on the two different passes. Probably not worth the effort in most cases, but maybe with certain pieces...

- I've found that noise reduction is utterly brilliant when it comes to removing JPG artifacts, but can have unexpectedly different effects (like sharpening, smoothing. light-balance) depending on the type of image.

- Also in terms of noise reduction, my experience is that using just the right amount (i.e. the Goldilocks principle) is the way to go. If you use too much, it can actually wind up removing detail it thinks is 'noise.'

- Other tools of course can compliment all this, like using "auto color-adjust," "sharpen" and other enhance tools in editor of choice. In the image steps I posted just above, I've edited in showing how I took that image back and forth.

Edit: With the caveat that some tools are best-used at different parts of the process. For example, enlarge should be used near the beginning, and sharpen near the end.
 
Last edited:
I tried the artwork option and not sure if it is better. I set it to the highest noise reduction at 4x.


Walking Dead, The ProX 1.08.png
 
In any case, all of these look better than the source, as it's much less blurry now. Good stuff.

I used Waifu again today to help me restore some 1963 arcade game parts. It's a non-pinball maybe possibly you remember as a little kid. Will post samples in a public thread.
 
Okay, I've been able to find enough answers that I've decided to move this thread in to public view. It's now a WIP guide, essentially.

So, a couple days ago I decided to try a new approach, that of blending two different source images together to create the best of both. In this case it was two photos of the same backglass, but at different angles and with different lighting, reflections and camera flash. The main benefit I was looking for here was to get rid of as much glare possible by sampling from the other image.

Preparation involved doing the usual unskewing and image upgrades, but also resizing both to match. Next, was importing them in to GIMP as layers. Next was adding an alpha channel to the layers, allowing you to delete sections from the top layer, creating holes through which the matching sections from the bottom layer appear.

In the third attached image, I'm done merging the two images, and am now starting to clean up problem areas by hand.

The point of choosing this particular backglass is that there's no finished-level versions available on the internet and in the pin-sim scene. So when this one's done, it should be a significant upgrade for the tables and B2S servers based on this machine.
 

Attachments

  • chicago_coin_moon_shot_bg1.jpg
    chicago_coin_moon_shot_bg1.jpg
    747.6 KB · Views: 45
  • chicago_coin_moon_shot_bg2.jpg
    chicago_coin_moon_shot_bg2.jpg
    775.2 KB · Views: 48
  • bg1-bg2.merged.retouching.jpg
    bg1-bg2.merged.retouching.jpg
    474.1 KB · Views: 41
Last edited:
I must say, it's a good feeling to start with a smallish source with lots of issues, quickly turning it in to something usable in-game.

That said, there's kind of dark flip side to that, in that the more I get better at this stuff, and the more of the bigger issues I clean up in an image, the more I see smaller issues still in need of work. Most of these probably don't really matter as part of a finished game, but it can be damn near impossible to unsee as the graphics repairer.

So a job which might take five minutes to repair 90% of the way towards a usable backglass might 'cry out' for a little more work to get it near-perfect, and before you know it, a couple hours have passed. Worse yet, looking at that 'finished product,' I can still see lots of problem that could use yet more work. It's a damnable rabbit hole, I tells ya.

OPP1.jpg
Lots of issues, but a nice source, with rich colors and pleasant lamp glow.

full

Twice as large now, with artifacts removed and most lighting
issues cleaned up. Still it needs... better sharpening, and stuff!


Fun detail: The original photog (Kevin McHugh, IPDB) captured a bright bulb in one of the ovals of the hanging art, top left. I thought it was a pretty good match, so edited around it.
 
As I was cleaning up the image below, I was once again confronted with the problem of: 'well, what about the fluffing reflection problems there, mate?' (see upper-left area)

(SOURCE)

I'd always sort of half-ass hoped that there'd be a proper anti-glare filter or 'heuristics heal' available, such that I could just lasso the whole problem area and 'fix the glare' with minimal fuss. But no.

An additional issue is that when trying to match a 'background fill area,' traditionally I've used lots of cloning in order to preserve a certain amount of texture. You know, the natural main area textures that almost any scanned / digitally photographed image has, even across areas with overall average, identical color.

What I dunce-head realised today, especially working with such simple artwork (and not photos), is that LAYERS are the way. I.e. when it comes to working on problem areas, stop trying to pixel-match, and instead think of the situation as multiple elements sitting on layers, which you'll eventually merge. So do break them up, go ahead and paint backgrounds without a care for texture, then drop the individual artwork back in. Haha, it works!

(latest WIP)
 
Nice work and progress Ike, though thinking about it, the layers idea is exactly how some of the artwork was made, multiple silk screens. Maybe that concept has a graphic equivalence, if you can separate each colour and assign it to a layer, then maybe...just thinking out loud..
 
Last edited:
Nice work and progress Ike, though thinking about it the layers idea is exactly how some of the artwork was made, multiple silk screens. Maybe that concept has a graphic equivalence, if you can separate each colour and assign it to a layer, then maybe...just thinking out loud..
Exactly, yes..

Yes, it seems like when you separate various graphics regions accordingly, that's when things seem to drop in to place, and that way you don't have to worry about 'slave-labor upon pixels,' so to speak.

I would think one of the magnificent examples of all this is (original, non-repair) is @shiva's JG art?
 
JG is a rather extreme example. With layers, you can group them into folders, and I have at least 30 folders with multiple layers in each folder. This allows you to move sections of the artwork around, and overlay over top of each other, as well as "hide" some folders (like the lens lights) while you are working on the art.

Even cleaned up, and removing any bits not used, there has to be close to 400 layers, and the psd file is just over 65 megs for a 4096x1873 size image.

Yes, you can separate colors, various ways to do it. Making a layer for each color, and using the magic wand tool to pick up the color is the one most people use. It's best to reduce the amount of colors first, (make as a "Indexed color" in ps) as anti-alias images (fuzzy lines) don't pick up well) JG was actually 64 colors total, as I use a color palette on a separate file I created.
 
Foof. So the more I work on this stuff, and as much as I've already improved the WIP guide in the first post, the more I realise just how much more there is to add, edit & correct. Ah well... but isn't that a good thing? :bonk:

Seriously-- it's downright amazing to me how a light touch with the right tools can produce credible results. I really am astonished on a regular basis, learning & realising what these tools can do with only a little experience. Okay then, I'll name the tools at the end and how I used them.

Let's start with PN's original image for A-B-C Bowler:

A-B-C Bowler - PN version.jpg

Anything's doable of course, but with a starter pic at 400x403, plus camera flash, the question is: how many hours of your time do you want to spend there, hoss?

Fear not, though. John Kirby uploaded a different shot to the IPDB, without camera flash:

A-B-C Bowler (Gottlieb, 1941) - IPDB.jpg

Yee-haw! :D
So then, the steps:

1) Waifu at high-setting noise reduction. Not too little, not too much so as to preserve as much detail as possible.
2) BigJPG to heuristically enlarge the image on the "art" setting, preserving clean lines.
3) GIMP's perspective tool to square the whole sucker up, meanwhile trying to offset image-bulging along the edges ("fisheye"). Note: I recommend an extra step in the first post if the image really needs to be perfect. Either way, it's much harder to line everything up when you have a double-border, as in this backglass.
4) Just because I'm silly, I made the whimsical choice to maintain the extra reddish hues along the S & SW borders, using the eye-dropper, bucket-fill & smudge tools to do the job.
Note: One really nice thing about the eye-dropper tool is that you can set it such that it picks an average sample of colors in the spread you choose (for example, one pixel, 3x3 of pixels, 72x72, whatever). This allows you to better match a bucket-fill region to surrounding areas, even if you need to 'smudge it up' afterwards in order to even things out. For example, in the past, I recall the eye-dropper sampling only a single pixel at a time, which could be downright maddening when trying to pick the best color from one dang pixel.
5) Applied FILTER/ENHANCE/SHARPEN at standard settings in order to make details pop and offset a certain residual blurriness. Note that GIMP's built-in sharpen settings are fairly aggressive, and that often it's good to tone them down a bit via the two sliders.

WIP (it's finished for now, honestly) below:

 
Last edited:
Forum activity
Help Users
You can interact with the ChatGPT Bot in any Chat Room and there is a dedicated room. The command is /ai followed by a space and then your ? or inquiry.
ie: /ai What is a EM Pinball Machine?
  • No one is chatting at the moment.
      Mibs Mibs: JonPurpleHaze has posted a new reply in the thread "What music are you listening to?".
      Back
      Top