Close
This post belongs to a parent post.
Search
Tags
  • ? anmi 661
  • ? feet 49840
  • ? garter 63257
  • ? no bra 189378
  • ? nopan 50715
  • ? open shirt 105407
  • ? underboob 14928
  • ? wet 78594
    • no pan no panties under boob soles nobra thigh band partially submerged underwater open kimono open clothes partially unbuttoned foot open cardigan swimming no pants huge feet foot focus open robe
Statistics
Options
Related Posts




anmi feet garter no_bra nopan open_shirt underboob wet

Edit | Respond

gerbil193
over 9 years ago
Somehow, the child post looks like it's got better quality, even if the resolution is lower. This one looks more blurred.
fireattack
over 9 years ago
Because that's a digital image.
yong
over 9 years ago
I could use some surface blur+sharpening filter, but it will also enhance the less visible paper texture.
This one is the fourth revision, previous version i processed years ago was applied with sharpening filter, but require extensive clean up, which is why i didn't post it until now.
Cyberbeing
over 9 years ago
In post #326118 I made an attempt at cleaning and matching the general color balance to the digital. I think it came out pretty decent, but at the very end I decided not to oversharpen it to the degree of the anmi pixiv digital to keep the edges free of sharpening halos.
yong
over 9 years ago
That's very nice indeed, but the filtering is tad a bit strong to my taste.
Amni's book has higher printing quality than average doujinshi, which probably can be done at 400dpi if that amount of noise filtering is acceptable.
I'm suppose i could rescan/reprocess her tea bag catalog 2, and recent Nardack scan which suffer the same problem.

I didn't know how much is too much, that's why some of my scans have coarse surface. /facepalm
Cyberbeing
over 9 years ago
anmi scans are always some of the the worse to filter because of that paper texture. You have very little options other than progressively applying somewhat strong filtering, restoring detail, and repeating until its fully clean. At the end that basically means going through with history brush and spot checking for any lost detail and edge sharpness.

Unfortunately it seems like I did overlook some highlight detail in a couple areas when I was restoring things, which does make it appear overfiltered. I went ahead and fixed those issues and uploaded a final filtered version in post #326303.

Overall I think it fine if you want to be conservative when filtering scans with a paper texture, since that is preferred to detail loss. It often takes way too much time to properly filter out a paper texture, sometimes requiring a rather tricky manual effort to get it looking good and even then having a moderate risk of failure if the texture is overlapping the details in the scan or just too intense. This one wasn't too bad since you had already gotten the texture moderately subdued.
yong
over 9 years ago
Their earlier doujinshis does not have paper texture, i didn't fully scanned their books because required to debind, and they are currently missing lol, gotta dig them out but it may need some times.

Are you using the CS6+G'MIC only combo? I'm using Paint.net for most of my scans, but it's either have limited options or I might need to change the way of my current scan processing workflow.
Some of my scans have jaggy lines which ordinary denoise filter cannot be filtered out, or the build-in resampler are the culprit.

Anyway thank for the assist, my scan probably would stay the same if without anyone help ;D
Cyberbeing
over 9 years ago
*Photoshop + 16bit color + GreycShop
*Waifu2x RGB NR2
*Darken Color blending for filtered layer /w tweaked opacity
(helps preserve lineart)
*Repeat above filter passes as needed
*Topaz Denoise5 (grain 0.10 for dithering shadows)
*Layer Masks for recovering highlight detail
*Patch tool (manual cleanup of anything filter passes missed)

This entire filtering job was primarily an experiment to see how well Waifu2x RGB NR2 (denoise only) would interact with GreyC when tackling this kind of problem. Overall positive in eliminating some of the hard to filter out distortions while also refining lineart, but it had a bad tendency of treating all highlight edges which weren't surrounded by lineart as banding and filtering them into smooth gradients. Nothing which layer masks couldn't resolve though.
fireattack
over 9 years ago
Cyberbeing said:
Photoshop + GreycShop + Waifu2x RGB NR2 + Topaz Denoise5
Darken Color blending for filtered layers @ various opacity
(helps preserve lineart)
Layer Masks for recovering highlight detail
Patch tool (manual cleanup of anything multiple filter passes had missed)

This entire filtering job was primarily an experiment to see how well Waifu2x RGB NR2 (denoise only) would interact with other filters when tackling this kind of problem. Overall positive in eliminating some of the hard to filter out distortions while also refining lineart, but it had a tendency of treating all highlight edges which weren't surrounded by lineart as banding and filtering them into smooth gradients. Nothing which layer masks couldn't resolve though.

Topaz Denoise was the only non-free filter, which I primarily use for dithering (grain 0.10) and cleaning up any banding in dark/shadow areas caused by greycstoration. I probably could have substituted non-free filters such as NeatImage or Noise Ninja as well if I desired a flat/smooth look in the shadows, but didn't this time because of what I was testing.
That's a lot of useful information, thanks for your effort.

By the way, won't Topaz Clean be better to smooth the dark area? Assuming you have access to this as well.

Another question, I have no idea what you mean by "NR2" of waifu2x, but anyway, which implementation do you recommend for local use (I have a nV graphic card if it matters)? caffe?
Cyberbeing
over 9 years ago
Topaz Denoise5 can be made to behave quite similarly to Clean3 in terms of smoothing dark areas, but I prefer Denoise5 overall since it allows me to reduce the overall noise level while adding back a lighter grain as dithering. Theoretically I could run both, but I usually don't.

"NR2" refers to "Noise Reduction Level 2".

Using either the tanakamura or caffe version is fine. Just avoid WL-Amigo waifu2x-converter-cpp which is bundled with koroshell ver1.01, since it doesn't support RGB filtering.