Moire samples on resizing.
It's scanned at 600dpi, then resized at 300dpi.
(PS)=Photoshop CS2, (Xn)=Xnview1.96
1) Bilinear also known as moire generator always fails :P
2) Bicubic/Lanczos have sharpaning effects, so they preserve screening, & cause a bit moire. btw, there is no visible difference between bicubic/lanczos at 50% resizing, but if it's resized at random parcentage, lanczos is a bit smooth.
3) Bicubic smoother/Mitchell haven't sharpening effects, then screening is a little reduced. they're very "safe" resizing imo.
4) Bspline gives out similar result to gaussian+resizing. it redusing any noizes while resizing, so you can use it when you'd like to avoid annoying editing. (and I'm using bspline the most :P)
It's scanned at 600dpi, then resized at 300dpi.
(PS)=Photoshop CS2, (Xn)=Xnview1.96
1) Bilinear also known as moire generator always fails :P
2) Bicubic/Lanczos have sharpaning effects, so they preserve screening, & cause a bit moire. btw, there is no visible difference between bicubic/lanczos at 50% resizing, but if it's resized at random parcentage, lanczos is a bit smooth.
3) Bicubic smoother/Mitchell haven't sharpening effects, then screening is a little reduced. they're very "safe" resizing imo.
4) Bspline gives out similar result to gaussian+resizing. it redusing any noizes while resizing, so you can use it when you'd like to avoid annoying editing. (and I'm using bspline the most :P)
So currently, what filters have you used other than resizing, Midzki?
Waiting for you comment ^^ (the text thingy)
Waiting for you comment ^^ (the text thingy)
Descreening with warpsharp
I was playing around random filters, and found Warpsharp script on Gimp2 works raw scans :)
Not similar to other sharpening filters, it can be used first pass to prevent over blurring edges.
I was playing around random filters, and found Warpsharp script on Gimp2 works raw scans :)
Not similar to other sharpening filters, it can be used first pass to prevent over blurring edges.
Darken only filtering
if you got annoyed by side effects of USM, just use "Darken only" layer mode.
how to: dupe layer > USM > set layermode darken.
it also works together with an edge detection mask.
it might also works on denoising, but I haven't managed effective way yet.
if you got annoyed by side effects of USM, just use "Darken only" layer mode.
how to: dupe layer > USM > set layermode darken.
it also works together with an edge detection mask.
it might also works on denoising, but I haven't managed effective way yet.
This is one image I've been fighting with for a while:
These "honeycomb" patterns are very hard to filter (in the dark blue, and less in the brown). Turning greyc way up looks like garbage. I can clean it up with a fairly heavy surface blur, but there's got to be a better way.
I tried warpsharp before greyc. It helps a bit, but since it's shrinking screening dots, it changes the color. The warpsharp version is noticably darker.
These "honeycomb" patterns are very hard to filter (in the dark blue, and less in the brown). Turning greyc way up looks like garbage. I can clean it up with a fairly heavy surface blur, but there's got to be a better way.
I tried warpsharp before greyc. It helps a bit, but since it's shrinking screening dots, it changes the color. The warpsharp version is noticably darker.
I'm also troubling filtering on the darks. I ended up keeping colors, and became using tone curves before/after filtering instead. even though, I couldn't keep contrast among edges/surface in the dark. still working on progress.
(blacking out the value of edges can avoid filtering, but I don't like this method :/)
to aoie: greyc, neatimage, surfaceblur, sgblur, warpsharp. and using them with masks (I'll write it later) :P
(blacking out the value of edges can avoid filtering, but I don't like this method :/)
to aoie: greyc, neatimage, surfaceblur, sgblur, warpsharp. and using them with masks (I'll write it later) :P
Same. Darker areas have bugged me so much. It's hard to filter them and preserve quality without destroying the area of it. For those parts I gotta do selective filtering. Healing Brush, lol ^^
I tried to suppress the intrusive color conamination with Noise ninja, turning luminance filtering strength to 0 and only let it deal with colors. Then two passes of greyc with different paramenters to eliminate luminance channel noise.
Then I tried to filter with neatimage on low and medium frequency range to filter out larger scale blockings, but this takes away too much detail and i gave up.
Tone mask.
Screening is proportional to the darkness level, therefore descreening strength should be adjusted to the tone of it's value. e.g. use 100% strength at darkest, 50% strength at midest, 0% strength at brightest.
How to: dupe layer > denoise > reveal mask & apply image from the layercopy with invert value
It can also helpfull to reduce scanning artifact/bleed through on the bright area. in this case, add mask without invertion, then manage tonecurves to get best ballance.
Screening is proportional to the darkness level, therefore descreening strength should be adjusted to the tone of it's value. e.g. use 100% strength at darkest, 50% strength at midest, 0% strength at brightest.
How to: dupe layer > denoise > reveal mask & apply image from the layercopy with invert value
It can also helpfull to reduce scanning artifact/bleed through on the bright area. in this case, add mask without invertion, then manage tonecurves to get best ballance.
Color adjustment: white point.
CG works are almost painted on true white canvas, hence white point must equal value 255. but the scans are not equal.
one reason is paper color. see forum #3137.
another reason is scanning/printing artifacts. due to that, even white borders aren't have same value. in this case, there are 3 ways to define white point.
1) take lowest value on white area. this will erase all noises on the white without filtering, but the result image is partly destroyed, and its contrast must get too high.
2) take midst value. since filtering mixes colors, it might be mostly used way. sometimes the result is bit high contrasted, so it might need curve adjustment.
3) take highest value. this will preserve original tone, but it requires lighten only filtering on the white area.
on the other hand, adjusting black point is more difficult. the darkest points of many original CG works are not value 0, and it has tough screening, and more, adjusting RGB in the mid tone also modifies RGB in the dark tone.
CG works are almost painted on true white canvas, hence white point must equal value 255. but the scans are not equal.
one reason is paper color. see forum #3137.
another reason is scanning/printing artifacts. due to that, even white borders aren't have same value. in this case, there are 3 ways to define white point.
1) take lowest value on white area. this will erase all noises on the white without filtering, but the result image is partly destroyed, and its contrast must get too high.
2) take midst value. since filtering mixes colors, it might be mostly used way. sometimes the result is bit high contrasted, so it might need curve adjustment.
3) take highest value. this will preserve original tone, but it requires lighten only filtering on the white area.
on the other hand, adjusting black point is more difficult. the darkest points of many original CG works are not value 0, and it has tough screening, and more, adjusting RGB in the mid tone also modifies RGB in the dark tone.
Is there someone always be able to make best profile on Neat Image? I couldn't :(
Since I failed making profiles many times, I gave up and now using only one profile sampled by a shadow area on AM screening.
I need to adjust Noise level (threshold) instead, however, better than annoyed getting different result each time. ofcause I can't use fine tuning, but I don't have any troubles since I started using tone mask..
I think noise detection of Neat image is very poor & rough, so I only use high-freq reduction to reduse chrominance only, or use it on a darken layer. use Greyc for other cases on reducing high-freq noise :)
Since I failed making profiles many times, I gave up and now using only one profile sampled by a shadow area on AM screening.
I need to adjust Noise level (threshold) instead, however, better than annoyed getting different result each time. ofcause I can't use fine tuning, but I don't have any troubles since I started using tone mask..
I think noise detection of Neat image is very poor & rough, so I only use high-freq reduction to reduse chrominance only, or use it on a darken layer. use Greyc for other cases on reducing high-freq noise :)
Don't just profile a single area, you need to complete your noise profile by fine-tuning (select a uniform area->right click->fine tune) with as many colors that you can find a uniform area in your image as possible to get a good result.
Thanks to your advice but still I coudn't get better result than using tone mask + tone curve. maybe I have to analyze the ranges of fine tuner and set them manually.
I also find it quite difficult to produce satisfactory results with neatimage against high-frequency AM Screening noise. It always tends to leave small spots of noise untouched...And turning noise threshold all the way up just blurs everything away. Also it often create ringing artifact along edges like greyc does.
I usually use the processing I posted at page 1 of this thread.
some precautions about noise ninja:
1.NN strongly desaturates thin colored lines such as blush lines (making them gray). So be sure to confirm these losses, and mask these areas if necessary.
Also, NN desaturates the whole image a bit.
2.NN sharpens result images by default, you'd surely like to turn sharpening off if NN is not the last step in your processing.
3.NN's profiler doesn't check selection area uniformity like Neatimage does, so don't mistakenly select an area that contains details.
I usually use the processing I posted at page 1 of this thread.
some precautions about noise ninja:
1.NN strongly desaturates thin colored lines such as blush lines (making them gray). So be sure to confirm these losses, and mask these areas if necessary.
Also, NN desaturates the whole image a bit.
2.NN sharpens result images by default, you'd surely like to turn sharpening off if NN is not the last step in your processing.
3.NN's profiler doesn't check selection area uniformity like Neatimage does, so don't mistakenly select an area that contains details.
PS Greyc doesn't cause ghosting.
The Alt Amp feature is pretty nice indeed to avoid that.
The GIMP Greyc seems to enjoy doing it to me.petopeto said:
PS Greyc doesn't cause ghosting.
'GREYCstoration is about to die'
So says the GREYCstoration site.
So says the GREYCstoration site.
Gah.. i might have to download and actually use gimp >.<
For what it's worth, I pointed out the fix to the GREYC author, but he never integrated it. (We didn't get off too well, so I didn't try hard to persuade him. I shouldn't have had to.)Radioactive said:
The GIMP Greyc seems to enjoy doing it to me.
GMIC looks like it takes a bunch of separate plugins, and just stuffs them into the same plugin. I can't imagine why you'd do that.Radioactive said:
'GREYCstoration is about to die'
So says the GREYCstoration site.
And it is still pretty buggy.petopeto said:
GMIC looks like it takes a bunch of separate plugins, and just stuffs them into the same plugin. I can't imagine why you'd do that.
(post #75708)
Since I saw this image, I noticed how NeatImage preserve the edges.
Adjust brightness of edges nearly equal 0, then apply NeatImage with setting lowest fine tuner into 0. then you can get similar result.
I think the reason is that noise level of edges become 0, hence they can avoid filtering.
this idea can also amplifiy on other filters.
for example: dupe layer > filtering > reveal mask > aplply image from back ground (under) layer into the mask > manage tonecurves at the mask.
but it have one problem. if edges couldn't reach at darkest value, it have to get filtering :/ (you can see this problem on the post #75708)
Won't you get ringing artifact with such aggressive profile settings? Most edegs have finite transition width, won't the filter screw up on these edges?
set aggressively, then soften parameters to reach at the looking normal.
the point is to know what's setting causes what result.
(although, I'm prefer into left screening than causing some ugly side effects.)
the point is to know what's setting causes what result.
(although, I'm prefer into left screening than causing some ugly side effects.)
I tried to resize to 50% with Xnview's bspline but I got extremely blurred results... Is this normal or my xnview is wrong?
Try to compare to irfanview's resizing.
btw, now I'm using gauss + lanczos resizing.
btw, now I'm using gauss + lanczos resizing.
The reason scanning 600dpi is to prevent moire effect, but FM screening never have moire.
The left of following image is 600dpi scan before resized, the right is 350dpi scan.
I doubt FM screening needs to scan over than final resolution.
The left of following image is 600dpi scan before resized, the right is 350dpi scan.
I doubt FM screening needs to scan over than final resolution.
on the other hand, AM screening needs to scan 600dpi
post #98993 300dpi scan has terrible moire. use it if you wanna troll somewhere
post #98995 400dpi scan hasn't remarkable moire, but it still there slightly
post #98997 600dpi scan hasn't moire anymore
post #98993 300dpi scan has terrible moire. use it if you wanna troll somewhere
post #98995 400dpi scan hasn't remarkable moire, but it still there slightly
post #98997 600dpi scan hasn't moire anymore
black out, white out, & selected surface blur for no gradation area
it's very simple technique, but very usefull :)
use tonecurves on the darkest/brightest areas, and use surface blur on mid tone areas
it's very simple technique, but very usefull :)
use tonecurves on the darkest/brightest areas, and use surface blur on mid tone areas
Pre-filtering on bright areas
Take a look at the raw (#1), the bright area on the image hasn't details, but there are ugly paper_texture & bleed_through.
I reduce them with a grayscale mask in the first editing process.
First, duplicate background layer then apply a filter to the whole image (#2)
Next, add a mask and apply back ground grayscale mask not to effect on shadow areas (#3)
Finally, lower gamma of the mask not to effect on mid tone areas (#4)
#5 shows only filtered area.
This masking technique is my basis, and I'm using this type of masks on the almost filtering process.
Take a look at the raw (#1), the bright area on the image hasn't details, but there are ugly paper_texture & bleed_through.
I reduce them with a grayscale mask in the first editing process.
First, duplicate background layer then apply a filter to the whole image (#2)
Next, add a mask and apply back ground grayscale mask not to effect on shadow areas (#3)
Finally, lower gamma of the mask not to effect on mid tone areas (#4)
#5 shows only filtered area.
This masking technique is my basis, and I'm using this type of masks on the almost filtering process.
midzki
Scanning techniques (misc)
see also image editing help