light washes out text AND noise in photoshop not in ST

Scan Tailor specific announcements, releases, workflows, tips, etc. NO FEATURE REQUESTS IN THIS FORUM, please.

Moderator: peterZ

dpc
Posts: 379
Joined: 01 Apr 2011, 18:05
Number of books owned: 0
Location: Issaquah, WA

Re: light washes out text AND noise in photoshop not in ST

Post by dpc »

Looking at the original image and tweaking with PS filters, it appears that you have two problems. The page underneath is bleeding through and the current page is being reflected back upon itself from what appears to be the adjacent platen pane or something else above.

As a test, take a black sheet of paper and put that over the adjacent platen pane so that any reflection off of that plane will be muted. Then put another black sheet under the page you're scanning to eliminate the bleed through. Alternating between these two methods as you tune your lighting and exposure levels could help in your attempts to determine what is the source of the reflection in your final image.

Dan had posted a good series of videos about light placement and platen angles. He demonstrated some of the problems you can see with platen planes that are 90 degrees apart. The new Hackerspace scanner design uses planes that are 100 degrees apart in an attempt to mitigate some of the inter-reflections between the two platen planes.
o3h1p
Posts: 71
Joined: 08 Nov 2010, 22:47

Re: light washes out text AND noise in photoshop not in ST

Post by o3h1p »

So apparently the scanner software settings are postscan changes, not input changes, so I can't see that being any better than photoshop. One other solution is to place a black sheet behind the page. This reduces the contrast from the others side and you have less noise. The only problem is that it makes a darker image. This does make it easier to clip out that particular color in photoshop or turn up the brightness, however.

Come to think of it I think this is a general problem with most books with sufficiently thin pages (probably most books out there), which is why I think you find most books on Google books have some of the light colors washed out from what I assume is a photoshop correction.

I'm becoming more convinced that this occurs with most thin-page books with color. Many textbooks are like this, which is what I mostly scan. I think most people here are doing thick paperbacks that are black and white so they don't notice this problem.

The question is with the right lighting/camera/settings/postprocessing can I correct for this noise without washing out other colors. If anyone has been able to scan a thin-paged color book without this 'noise' problem, please let me know.
User avatar
Heelgrasper
Posts: 70
Joined: 19 Feb 2012, 21:04
E-book readers owned: None
Number of books owned: 500
Location: Randers, Denmark

Re: light washes out text AND noise in photoshop not in ST

Post by Heelgrasper »

dpc wrote:Then put another black sheet under the page you're scanning to eliminate the bleed through.
That isn't going to eliminate the bleed through since that comes from the print on the other side of the same sheet.
---
Jakob Øhlenschlæger
Randers, Denmark

The past is a foreign country: they do things differently there
L. P. Hartley
o3h1p
Posts: 71
Joined: 08 Nov 2010, 22:47

Re: light washes out text AND noise in photoshop not in ST

Post by o3h1p »

@dpc Yes I did the black paper test (with a single page on a flatbed) and it does help equalize the bleed-through so that it's easier to remove post processing. The problem with this is that it would be a lot of work to move two sheets of paper around for each page. I'm still hoping that with the right settings/lighting I can get better output. But it might just be the case that any sufficient bright light will make it bleed through.
dpc
Posts: 379
Joined: 01 Apr 2011, 18:05
Number of books owned: 0
Location: Issaquah, WA

Re: light washes out text AND noise in photoshop not in ST

Post by dpc »

o3h1p wrote:@dpc Yes I did the black paper test (with a single page on a flatbed) and it does help equalize the bleed-through so that it's easier to remove post processing. The problem with this is that it would be a lot of work to move two sheets of paper around for each page. I'm still hoping that with the right settings/lighting I can get better output. But it might just be the case that any sufficient bright light will make it bleed through.
Right. The black paper is just to help isolate the noise a bit for you to determine where it's coming from and then adjust your camera settings or lighting to try to mitigate the problem.
heelgrasper wrote:That isn't going to eliminate the bleed through since that comes from the print on the other side of the same sheet.
You're right. If it's true ink bleed through to the front side of the page, the black paper trick won't make a difference. There are certain instances where it does help though. My old flatbed HP scanner has a bright white pad under the lid and when I scan documents that have printing on both sides the print on the back can show through. If I put black paper over that white pad, the text on the back side is no longer visible.

One idea that I've been toying with is to do a post-process operation with the a scanned page, and the image from its back side. You'd perform a blend operation between the two pages that would subtract a percentage of what is on the back side image from what is on the front. Proper registration of the two images is a bear (they come from different cameras), but the early results show some promise.
o3h1p
Posts: 71
Joined: 08 Nov 2010, 22:47

Re: light washes out text AND noise in photoshop not in ST

Post by o3h1p »

dpc wrote:One idea that I've been toying with is to do a post-process operation with the a scanned page, and the image from its back side. You'd perform a blend operation between the two pages that would subtract a percentage of what is on the back side image from what is on the front. Proper registration of the two images is a bear (they come from different cameras), but the early results show some promise.
Now this is interesting. Unfortunately I'm not sure how you would do this to two pages, let alone in batch (with photoshop). Let me know if you get it working.
spamsickle
Posts: 596
Joined: 06 Jun 2009, 23:57

Re: light washes out text AND noise in photoshop not in ST

Post by spamsickle »

The washed-out text is (IMO) due to reflection/glare from the light which also appears in the photograph. If you adjust your lights or cameras so that light doesn't appear, it should resolve the washed-out text problem.

I would say that the black piece of paper under the page you're shooting would be the correct solution to the ghosting problem, rather than attempting to subtract the ghost image in software. Aside from registration problems, you're going to have to consider the front image as well - the back-page image will not show through dark areas of the front image, so simply subtracting a percentage will not work. The back-page image is not showing because the ink bled through the page, but because the light shone through the page and bounced back. Letting it bounce off a uniform black surface should eliminate virtually all of the back-page ghosting.
o3h1p
Posts: 71
Joined: 08 Nov 2010, 22:47

Re: light washes out text AND noise in photoshop not in ST

Post by o3h1p »

spamsickle wrote:I would say that the black piece of paper under the page you're shooting would be the correct solution to the ghosting problem, rather than attempting to subtract the ghost image in software. Aside from registration problems, you're going to have to consider the front image as well - the back-page image will not show through dark areas of the front image, so simply subtracting a percentage will not work. The back-page image is not showing because the ink bled through the page, but because the light shone through the page and bounced back. Letting it bounce off a uniform black surface should eliminate virtually all of the back-page ghosting.
I wouldn't say 'eliminate' all the back-page ghosting but rather removing any variations in the hosting. You still get the reflection phenomenon that you are referring too, but it occurs for the *entire* page. This makes it more uniform, but gives the page a black-blueish hue (think darker). The page is no longer white and the problem is so obvious that pasting a little white area in the margin shows the difference is stark--it certainly isn't a white page anymore. So it turns out that if you want that white page (I do) you need to run through some post processing still. There are many ways, these seem to be the most successful:

1) Mess with the levels -- if you clip the high end you get rid of this color, but unfortunately you wash out some of the light colors
2) Use the magic wand tool and cut. This is a decent solution but the problem is that it misses the center of enclosed letters like "o" or "e" and you get the dark hue inside
3) Select by color and cut -- this is where the black-paper method shines, now that you have a uniform hue (since there are no changes in contrast or different colors causing the reflection-- everything is the same black) you can select this color and cut it out. The problem occurs when this color appears on the page that you cut out. Also there is a logistical problem of placing black pages between every page.

There is just no ideal solution that I can find with these full-color books. I'm still hoping with the right camera settings you can get the equivalent of a 'black-paper-behind-the-page' shot without actually moving two black pages around (which increases the scanning time by at least a factor of 2 or 3.
spamsickle
Posts: 596
Joined: 06 Jun 2009, 23:57

Re: light washes out text AND noise in photoshop not in ST

Post by spamsickle »

A white page will appear gray if it's photographed with a short exposure, and a gray page will appear white if it's photographed with a long exposure. You are correct that a black-backed page will not appear as bright as a random-page-backed page, but as you note it does equalize the background and eliminate the variation caused by the printing on the back of the page. At that point, getting a picture in which the page is white, or the text is black, or something in between, should just be a matter of adjusting the exposure. You could even use CHDK to bracket your exposures, and take three (or more) shots per page instead of only one.

While shuffling a black page in and out will increase the shooting time, I shoot a leisurely ten shots (20 pages) per minute now. If I triple the time, and drop to 3 shots (6 pages) per minute, that's still 350 pages per hour. When I think of your proposed software post-processing solutions, it's hard for me to imagine (for instance) that doing a magic-wand selection and waiting for Photoshop to load, process, and save an image isn't going to eat up much more than the six seconds per page you might have spent solving the problem on the front end. I can't imagine a post-processing software solution which will be robust enough to script, so it will be your time rather than the computer's time that's consumed.
o3h1p
Posts: 71
Joined: 08 Nov 2010, 22:47

Re: light washes out text AND noise in photoshop not in ST

Post by o3h1p »

I have also tried adjusting exposure, offset, gamma to fix shadow (in photoshop, not the camera) and it also had some problems. I haven't tried it with the black page method yet though, so I'll have to try it and report results. I'm not sure if its the best solution yet. The problem is the darker background affects the different colors differently. The whites are darkened the most and the blacks are darkened the least, what ends up happening when you adjust the exposure is that you start to wash out the darks. (Though I could have sworn that the bigger problem was the loss of the light colors not the wash out of the dark colors, though I think I noticed both.) I'm going to try a bit of experimenting next weekend and I'll report back results.

In regards to the logistics issue: I currently run at >1000 pages an hour and I (personally) consider a reduction to 30% output a big effect when I spend an entire day going through thousands of pages. Also it is a pain having to remove the black page from behind, turn two pages, place the black page, and turn back all while holding the platen up with your other hand. However it turns out that photoshop is highly 'scriptable' and a magic wand selection (for instance; I still think select-by-color is the best) can run over 300 pages in maybe about 10 min which is fully automated (the 'computer's time' not yours), so you hit a single button and you go and scan the next book. But it's not the automation method that is the issue (photoshop's 'scripting' is sufficiently robust) the issue is that none of the solutions that I gave is all that great even if you did 'script' them. The black page method certainly does improve results, but you take a pretty big hit in speed (for me, at least) and the results have their own set of problems.
Post Reply