Close
what's your processing?
I think we should exchange our own processing to improve quality of scans.
Let us show your processing (and forgive me my poor English ^^)!

For me,

1. adjust only color levels manually (not use auto levels) while a scanner shows a preview, and make other filters all turnd off.
2. scan all with 600dpi.
3. if scans have moire, resize them into 300~450dpi.
4. apply Greycstortion twice with different settings.
1st setting is p=0.7~0.8, a=0.2, α=1.0~1.5, σ=5.0
2nd setting is p=0.4~0.6, a=0.4~0.6, α=0, σ=2.5
(this settings may give you overfiltered images)
5. finally, open them with a paint tool like Opencanvas or SAI, and repaint them to erase some dusts, creases, and characters.

thanks ^^
mine is:

->scan at 1200 dpi
-> fix eventual crease/dust/
-> resize to 600 dpi
-> fix the the stuff left or merge in case of 2pages scan
-> Noise ninja (12-12) setting
-> resize to 300 dpi
-> very low Noise ninja again (10-10)
-> auto-contrast to fix brightness/contrast levels
-> save
here is mine (I have little experience though)

This method is largely based on this one:
http://www.pireze.org/blog/?p=250

10->All filters off, scanner sensors set to full range at 600dpi/24bit
20->noise ninja (just make sure at least no noise is amplified--this occassionally appears with bad noise profiling), save this.
30->smart blur, while preserving as much detail as possible to eliminate local color variations.
40->Dust&scratches at 1px 1 or 2 passes, to eliminate dots mistakenly left out by the step above.
50->Inspect the document and see if I have lost too much detail-- if yes, back to 30 and use a milder setting.
60->Remove crease and other artifacts--I do this last because scanning moire sometimes prevent me from correctly determining colors.
70->See if there are portions where I want detail over low noise, load the mid-product saved in 20 and replace them.
80->adjust the colors, trim out borders, correct roll error.
90->resize to 300 dpi.save.

also, apology for my poor english ^^
I recommend posting a raw image and the same scanned image to the site, and linking them from here. This way, people can easily see "before/after".

midzki said:
1. adjust only color levels manually (not use auto levels) while a scanner shows a preview, and make other filters all turnd off.
Levelling by hand can be very difficult. Using an IT8.7/2 target (such as from http://www.targets.coloraid.de, R1, $20 shipped) can do this accurately, giving exact colors instead of having to do it by eye. (If you have a good eye, it might not be necessary, but it's cheap and also eliminates errors due to monitor error.)

2. scan all with 600dpi.
3. if scans have moire, resize them into 300~450dpi.
If you have moire at 600 DPI, resizing down is going to blur the screening away (poorly) and may make it harder to get good greycstoration output. I'd try leaving it at 600 DPI and increasing alpha in greycstoration a bit, maybe...

(The real fix is to scan at a higher DPI, but that can be very slow.)

4. apply Greycstortion twice with different settings.
1st setting is p=0.7~0.8, a=0.2, α=1.0~1.5, σ=5.0
2nd setting is p=0.4~0.6, a=0.4~0.6, α=0, σ=2.5
(this settings may give you overfiltered images)
Greyc settings used at 600 DPI will blur the image more at 300 DPI. You may want to lower alpha a bit for 300 or 450 DPI compared to 600 DPI.

70->See if there are portions where I want detail over low noise, load the mid-product saved in 20 and replace them.
For images I'm giving more attention to, I sometimes do this:

- load into Photoshop (this is where I do all my greyc work anyway)
- duplicate the layer (control J)
- select the top layer, and run Greyc normally
- select the bottom layer, and run Greyc with lower settings appropriate to the smaller, detailed parts of the image (blush lines are a very common problem)
- select eraser tool, set a low hardness (~25%), and erase from the top layer where I want finer detail
petopeto said:
- select eraser tool, set a low hardness (~25%), and erase from the top layer where I want finer detail
The better method would be to just make a layer mask (non-destructive) and mask out the areas on the top level you want to keep detailed with your brush opacity of choice. This way you won't have to abuse your history states if you make a mistake with the eraser (destructive), and makes it easier to tweak how much detail you want retained if you change your mind late in the process.
petopeto said:
I recommend posting a raw image and the same scanned image to the site, and linking them from here. This way, people can easily see "before/after".
post #51908 is 600dpi raw.
post #51909 is 300dpi after processed.

petopeto said it "tricky",
yes, I totally agree :p

some details such as blush lines, and patterns on the background were destroyed, but still it looks not ugly.
It doesn't look ugly, but I think it can be improved. It loses enough detail that the difference is visible even in the 1363x1000 sample image.

I'm playing around with it in greyc to see if I can do better, but I don't have any special talent here. If anyone wants to give a shot: try taking post #51908 and filtering it so the noise is minimized, but the fine details remain (watch the background pattern, the wood grain underneith, the small highlight on her eyes, and the highlights on the water drops). Don't crop the image, so any variants can be compared right on top of each other.

One attempt in post #51914: just a simple greyc pass (amp 80, sharpness .6, aniso .3, alpha 2, gfact 1.5) and then resized down to 2303x1689. More noise is visible, but it removes enough so it looks more like texture than an artifact, and more details are preserved.
My attempt v2: post #52232
Deleted first attempt (post #51966) for the issues midzki noted.
petopeto said:
Levelling by hand can be very difficult. Using an IT8.7/2 target (such as from http://www.targets.coloraid.de, R1, $20 shipped) can do this accurately, giving exact colors instead of having to do it by eye. (If you have a good eye, it might not be necessary, but it's cheap and also eliminates errors due to monitor error.)
haha, I simply give up to adjust colors :p
My scanner, and sRGBprofile didn't work well. so I adjust white/black point only, didn't touch gammas,curves,etc.
Maybe gammas of my scans are a bit high, so they must need to be fixed.

Cyberbeing said:
My attempt: post #51966
It seems to be applyed something like color convert or tone curve, isn't it?
Maybe just stick to descreening for that image, to see what methods give the best noise/detail balance (and describe what you do, so it's reproducable).
midzki said:
It seems to be applyed something like color convert or tone curve, isn't it?
It seems NoiseNinja caused this problem when I used it to reduce color noise. I guess that means I shouldn't use NoiseNinja for this type of image.

petopeto said:
Maybe just stick to descreening for that image, to see what methods give the best noise/detail balance (and describe what you do, so it's reproducable).
I've redone the image from scratch with a different method and copied down the filtering steps I took this time (end result had around 10 layers). My goal was to slowly etch away the noise to make it as soft as possible, while keeping detail loss to a bare minimum along the way as well as trying not to mess up the color. Here is the end result: post #52232

Median 2px (25% Fill) << Begin to reduce screening. I set 25% Fill because I it seemed like any stronger killed too much detail.

Reduce Noise Str 0 | 25% Color | 0% Sharp (25% Fill) << Light color noise removal. Even though I had Reduce Noise set to color only, it still touches luminance, so I wanted to have it filtered a bit.

Reduce Noise Str 0 | 25% Color | 0% Sharp (75% Fill/Color Blend) << Heavier color noise removal on color only (this time no luminance is touched)

Bilinear Resize to 300dpi << I decided to resize at this point because Despeckle did better on the smaller version

Despeckle (50% Fill) << removes bright specs

Add noise 2% uniform monochrome (75% Fill) << Creates some dither and brings back some perceived detail.

Neat Image (0,0,0|0,0,0|100,25,25|25,75,25|HQ,HR,SE|Sharp Disabled) << Remove some of the noise you created while retaining most of the dither

Reduce Noise Str 1 | 60% Detail | 25% Color | 0% Sharp (Opacity 75%/Fill 75%) << Remove a bit more luminance and color noise with high detail retention

Copy and Combine Layers << At this point I am done with the main noise reduction so I just want to work from the final product so far.

Create Layer Mask with girl and smooth (untextured) areas exposed << I don't want to filter the wood and detailed areas anymore

Gaussian Blur Layer Mask 3px << So the transition between your filtered and unfiltered areas is smoother.

Greysctoration 30 0.5 1.0 0.3 3 0.8 30 2 2 near alt (50% Fill) << Smoothing the girl and smooth areas with a somewhat strong/fast filtering

High Pass Sharpen (Radius 2, 30% Opacity) << I went with a minimal opacity value on the high pass layer, but it could easily have been set higher to increase sharpness/contrast even more. << This brings some contrast and sharpness back to the image which was lost from the previous filtering. If you don't know what high pass sharpening is: http://www.luminous-landscape.com/tutorials/high-pass-sharpening.shtml

Clone stamp to clean up the bright areas in the bottom left corner that didn't look like they should be there. << Manual work

I chose to use fill over opacity most of the time because I liked the effect better. The values for each were found by a lot of tweaking and experimentation to see what looked best.
__________
Note: I've left out some steps where I created layers and what I created them from. Unfortunately, I didn't write those things down, but you should be able to figure it out with a little common sense. It would usually be either a copy of the result of top layer or of the previous layer. This would involve merging some layers since I had fill levels set.
No wonder my scan ssucks compared to ya guys...

Welll at least i can take these to heart and use them for my next scans.
Cyberbeing said:
It seems NoiseNinja caused this problem when I used it to reduce color noise. I guess that means I shouldn't use NoiseNinja for this type of image.
I didn't know these kind of filters which causes too much color shift, so I wanted to know what you used.
if Noise ninja causes always this color shift, I never use it.

Cyberbeing said:
I've redone the image from scratch with a different method and copied down the filtering steps I took this time (end result had around 10 layers). My goal was to slowly etch away the noise to make it as soft as possible, while keeping detail loss to a bare minimum along the way as well as trying not to mess up the color. Here is the end result: post #52232
I hardly understand this process because of my knowledge of English(>_<) and Photoshop (I use Gimp), but the result is well details preserved in my eyes.

Still I'm testing another approach.
My goal is a bit different from Cyberbeing. my priority is reduse noise perfectly, and preserve details comes next.
midzki, since you use Gimp, what do you use for noise reduction?
Cyberbeing said:
midzki, since you use Gimp, what do you use for noise reduction?
I use mainly Greycstoration, and sometimes, use SGBlur, WaveletDenoise, and UnsharpMask2.
once I have used despeckle, but it couldn't give out good result.

and often use paint tools instead of filters ^^
(but I want to know the way which doesn't need paint tools)
btw, my 2nd attempt
post #52527

It's very simple, but if it comes to imeges which have no complicated patterns like megami mags, it works quite well.

1) make a mask
dupulicate layer, sharpen(+50), edge detection(max), desaturate(average), e.t.c.(sorry, I didn't remenber my whole processing >_<)

2) apply Greyc (dt60, p0.3 a0.5, α0.6, σ4.0, fast off) only edges

3) then apply Greyc(60, 0.7, 0.4, 0.6, 2.0) again onto whole the image.
In order for this to be generally useful, let's stick to techniques that are generally usable. If you use several passes on an image, try to explain why you did it, so it's easy to understand and apply to other situations.

Post #52817 (original: post #52818):

First, reduce the large screening. Greyc at dt 40, p .5, a .4, alpha 1.3. (The high alpha is to compensate for the "large" screening.) This spreads out the larger noise without reducing detail.

Second, reduce the smaller noise that the last pass left. Greyc again at p .6, a .4, alpha .5. (The small alpha is because we're now dealing with fine noise.)

This does a pretty good job, but there's still noise clearly visible in the darker areas. I tried something different here: select just the darker areas (Select -> Color range -> Shadows) and run greyc again. This time, sharpness .8, aniso .6, alpha .5. (Still low alpha since we're again dealing with fine noise; raise the sharpness or we'll start to damage details.)

Then standard stuff: rotated a little to fix alignment, reduced to 65%, and cropped.

It could probably be improved a bit, but it looks decent, and this is all scriptable (except for the final rotation/crop), so I can make an action and apply it to the rest of the images in this set automatically--which is important; I usually don't want to spend hours fiddling with dozens of images one at a time.

Note that when I did each individual part, I saved it to a Photoshop action. Then, expanding the action shows the settings, and I can drag them together into a single action when I'm done for batching.
Post #52994 (original also post #52818):

The irregular noise that was left behind in post #52817 bugs me: it looks more like JPEG artifacts than anything else. Tried replacing the shadow blur with a regular surface blur (rad 5, thr 8). It seems to work a bit better: the darks aren't blurred quite as much, but it's more consistent in the lighter areas.

Both of these completely lose the blush lines. They're just so fine that I don't know anything that can be done about it. midzki: are the blush lines visible on the actual page? I've seen some POP images where blush lines are visible in the scan that I can't see on paper.
petopeto said:
midzki: are the blush lines visible on the actual page? I've seen some POP images where blush lines are visible in the scan that I can't see on paper.
I can see clearly not only blush lines but some moire on the actual paper ^^
but I think almost every people couldn't recognize it, and never care (ofcause I can recognize it , but I don't care because I have no idea which can preserve details perfectly not using paint tools ^^).
midzki said:
if Noise ninja causes always this color shift, I never use it.
It doesn't seem to be the case, noise ninja works nice for me... Or maybe it goes wrong on some special occassions, so it would be nice to compare the colors before further proceeding.
quoted from my another post

midzki said:
btw, I found filtering Greycstoration 2times with completely defferent parameters is really effective.

for example,
1st(-dt 10 -p 0.2 -a 0.6 -alpha 3 -sigma 0 -fast false)
2nd(-dt 30 -p 0.2 -a 1.0 -alpha 0 -sigma 3 -fast false)

1st setting is reducing moire, & 2nd setting is smoothing edges.
The point is setting -alpha and -sigma separately.
I'll show the example image later.
This is the result. it still 600dpi no size down.
post #56731

It was applied only Greycstoration, so no need to use photoshop or gimp.
I think it's really easy & fast method.
Are you sure they weren't both -a 0.6? 1.0 tends to cause swirly artifacts, but I don't see that. (Is this a different scan? The colors are different and it looks a little squished horizontally.)

It looks a little funny up close, wit those straight lines near edges.

-a 0.2 -alpha 1 -sigma 1 -gauss 1 -fast -alt (one pass) doesn't do badly. Loses more of the blush than I'd like (would probably do a separate mask pass for that).
petopeto said:
Are you sure they weren't both -a 0.6? 1.0 tends to cause swirly artifacts, but I don't see that.
Interesting?
I set parameters extremely to show how -a & -sigma works.

[edited]

let's me try more sophisticated way with the old scan.
post #56929
1) gaussian blur: 2px
2) greycstoration: -dt 10 -p 0.2 -a 0.6 -alpha 3.0 -sigma 0
3) greycstoration: -dt 20 -p 0.2 -a 0.7 -alpha 1.0 -sigma 3.0
4) high pass sharpen: gauss 5px opacity 60%
On high-res scans (600 dpi or more), I find it quite useful to eliminate coarse color noise with a denoising soft based on noise profiling such as noise ninja before finishing-up with greyc. These are quite effective against images with large uniform areas.
These softwares sometimes cause minor color shifts due to their nature though, so be careful.

IMO, for a quick comparison, noise ninja is based purely on noise profiling and employs little or no structure detection, while Neatimage employs both. And greyc doesn't have profiling and relies solely on blurring along structure vectors. I got to these conclusions by inspecting their resulting artifacts, and have no proof so don’t take it too seriously.

Note that it doesn’t imply that I think Neatimage does better job. In fact I use noiseninja + greyc myself.

Also something off topic: In fact all these methods eliminate noise by simply averaging, which is the only method against random ISO noise produced by CCD sensors. In fact CMYK screening is not random, it should be possible to solve back to a much better source image if we take the periodicity of screening into account. But it seems no easy job, maybe enough for a master degree research project… Would a nice lab please have someone work on this?....
Guess I'll share mine.

1. Use the Scanning Software for initial screening. Which is the Epson Scan. (this removes all of the morie effects, so i don't have to worry about it)

  • Scan at 450 - 600dpi, I usually scan at 500dpi at 24bit. As I don't resize my scans later on.
2. With my Photoshop CS2

  • Crop the scan
  • Rotate it if necessary (as most scans are .010% - .050% off)
  • Aline other scans if it's a multipage spread or scan
and the Neat Image PS Plugin I use the settings.

  • Profile your image first, you'll need to find an area at least 128x128, anything lower and your profiling will be worsen.
Noise Level -
  • Luminance Channel - 80%
  • Chrominance Channel - 120%
Noise Reduction Amount -
  • Luminance Channel - 80%
  • Chrominance Channel - 90%
  • Click on Smooth Edges. Very Low Frequency for paper textures.
  • Sharpening at 0% ( I don't like this feature, as I can sharpen it with PS tools)
~~ Usually a second pass will clear it up more, so try it again with these settings.

Noise Level -
  • Luminance Channel - 40-50%
  • Chrominance Channel - 0%
Noise Reduction Amount -
  • Luminance Channel - 40-50%
  • Chrominance Channel - 0%
œ Chromaniance Channel isn't needed no more as you have filtered out most to all of the noise. Specifically the green, blue and yellow crap.

3. With paper textures on darker areas, create a layer mask for it with either the quick mask or select it with the polygone lasso or the pen tool.

  • Guassian Blur at 1-2 pixel or median for rougher surfaces scans.
  • You can use Dust and Scratch at 1 - 3 pixels using the preview to view how it will looking like, as this techniques has sometimes helped for me.
  • I don't like despeckle as you can't control the effects, and it never works for me.
  • I use the same settings for paper textures, Altering the Luminance Channel usually from 70 - 95%, never 100%. At 100, you're basically blurring the whole image.
  • Usually by this time my scan is a bit blurred, and i use Unmask Sharp (not peto's addon, as i haven't downloaded it yet)
œ As a note, Noise Ninja and Neat Image screening does poorly on corners

-Settings

  • Amount , 50 -100% depending on how badly it needs it.
  • Radius , 1.0 - 1.5 pixels
  • Threshold , 0
  • If this isn't cleaning it up, duplicate the layer. Repeat step 2 again with the duplicate layer with much stronger settings this time and set the layer above it and change the transparency to around 10-15%, altering this to the appropriate level to gain the best result.
4. At this point i'll have any scans filtered and screened out, and i'll do color adjustment here.

  • I've found the best results is a mix b/t using the Level Tool, the the Color balance and vary amount of similar color with the brush tool.
  • Usually with the scan, brightness is lost so i add a Curve Layer. I don't necessarily use a specific setting, just use it to the best of the original as possible.
5.

  • Look and see if it matches that of the physical copy and save it as PNG none interlaced and i'm done.
ps: My greycstoration doesn't seem to work at all with any settings other than initial blur that makes it goes all black so I didn't add it in.
Thanks a lot petopeto, been using your action and it works very well.

petopeto said:
This does a pretty good job, but there's still noise clearly visible in the darker areas. I tried something different here: select just the darker areas (Select -> Color range -> Shadows) and run greyc again.
Actually, there's a way to get a more refined selection.

1) Duplicate the layer, double click on it to bring up blending options. It will bring up a dialog box.

2) Look for the "Blend if: Gray" and the "This layer:" slider and grab the white triangle handle and adjust it downwards towards black(alt click and drag the triangle to split it in half). Drag the split triangle to adjust blending. I found that settings of 0(black triangle), 100(left-half of white tri) and 180(right-half of white tri) work well.

3) Run Greycstoration on the layer.

4) Flatten image.

This method gives you a very smooth result and more control over the darks that receive the noise reduction.
it's interesting to read peoples' techniques for processing scans, interesting and informative...
of course i was actually tempted to just type in 'core i7 920' just to try to be funny.
still, interesting stuff, lots of techniques worth keeping in mind if i ever start scanning