If I save that image (post #61237) as max quality JPEG (quality 12 optimized) in PS, it's 1025k (40% smaller), and I have to zoom in very far to see any difference.
PS cs2 level 12 (1056k) ≒ 99%
PS cs2 level 11 (678k) ≒ 95%
I guess.
jpeg compression of level 11 got visible artifacts with enlarged view, and level 10 is visible with normal view.
but everyone love to use level 10 or under :|
PS cs2 level 11 (678k) ≒ 95%
I guess.
jpeg compression of level 11 got visible artifacts with enlarged view, and level 10 is visible with normal view.
but everyone love to use level 10 or under :|
Suemura uses level 5-6 i think :/
Examined.
He doesn't use PS, but compression maybe 85% ≒ level 9.
To be honest, they're ugly and eye sore (-ω-)
ps: MS paint may use down sampled 75%. it's above human's native sense.
He doesn't use PS, but compression maybe 85% ≒ level 9.
To be honest, they're ugly and eye sore (-ω-)
ps: MS paint may use down sampled 75%. it's above human's native sense.
Note that different JPEG software will use different levels of chroma subsampling, independent of the quality setting.
XnView defaults to 4:2:0 subsampling, which means color resolution (the UV part of YUV) is halved in each dimension. You can change this to 4:4:4 in the compression options (it calls it "1x1, 1x1, 1x1" instead of the name everyone else uses, 4:4:4), which will disable subsampling. This means that XnView JPEG is lower quality, even if you raise the quality slider, unless you know to fix the subsampling.
Google says Photoshop doesn't use chroma subsampling above quality 7 (this may vary across versions). I've confirmed this at 11-12.
XnView defaults to 4:2:0 subsampling, which means color resolution (the UV part of YUV) is halved in each dimension. You can change this to 4:4:4 in the compression options (it calls it "1x1, 1x1, 1x1" instead of the name everyone else uses, 4:4:4), which will disable subsampling. This means that XnView JPEG is lower quality, even if you raise the quality slider, unless you know to fix the subsampling.
Google says Photoshop doesn't use chroma subsampling above quality 7 (this may vary across versions). I've confirmed this at 11-12.
Gimp also set 4:1:1 as default. need to change 1:1:1 first.
Irfanview doesn't have subsampling option. don't use it.
misc about jpeg setting:
・use always DCT:float, Huffman optimized.
・removing a thumb and other metadeata decreases file size a bit
・progressive make files slower to open. even though it decreases file size a bit, don't use it.
Irfanview doesn't have subsampling option. don't use it.
misc about jpeg setting:
・use always DCT:float, Huffman optimized.
・removing a thumb and other metadeata decreases file size a bit
・progressive make files slower to open. even though it decreases file size a bit, don't use it.
"Progressive" for JPEGs means it's the sort that opens in several passes. It's designed for modems (the extra-slow sort, eg. 2400 baud), and it looks terrible (everything is blurry/blocky until it finishes loading).
(The rough PNG equivalent of this is called "interlaced". Don't use that, either. Don't get these mixed up with video, where "progressive" and "interlaced" are opposites.)
(The rough PNG equivalent of this is called "interlaced". Don't use that, either. Don't get these mixed up with video, where "progressive" and "interlaced" are opposites.)
These is ACDSee's config dialog.
Do you mean I should uncheck any one in "Color component sampling" to get the better quality? >_>
Do you mean I should uncheck any one in "Color component sampling" to get the better quality? >_>
Can you see desaturation on lips & blush lines?
#1 is png. #2 is dawn sampled jpeg 100%.
There is always visible losses on down sampling.
don't use it.
check both may mean 4x4:1x1:1x1. uncheck themfireattack said:
These is ACDSee's config dialog.
Do you mean I should uncheck any one in "Color component sampling" to get the better quality? >_>
A CCD scanner scans with a resolution of integral dividing part of max optical resolution.
if your scanner has 2400 max dpi optical resolution, the scanner can scan 1200, 800, 600, 480, 400, 300, e.t.c. (or just only 1200, 600, 300)
if you scan at random resolution, the scanner resizes images after scanned.
if your scanner has 2400 max dpi optical resolution, the scanner can scan 1200, 800, 600, 480, 400, 300, e.t.c. (or just only 1200, 600, 300)
if you scan at random resolution, the scanner resizes images after scanned.
All CCD scanners always has distortion due to their mechanism.
for example, my gt x820 has a distortion following the image
this scanner 's distortion must be minimum, since my old scanners have huge visible distortions. e.x. gt s600 has 5% width extention than height. canoscan 5600f has parallelogram distortion.
some distortions might be fixable, so measure your scanner to get better result.
for example, my gt x820 has a distortion following the image
this scanner 's distortion must be minimum, since my old scanners have huge visible distortions. e.x. gt s600 has 5% width extention than height. canoscan 5600f has parallelogram distortion.
some distortions might be fixable, so measure your scanner to get better result.
There are some techniques make edge mask more neatly.
1) before applying edge detection, raise gamma. it masking more darks, and less brights.
2) apply surface blur after edge detected. it makes surfaces easily mask out with leveling.
3) of cause, leveling is the most important part of edge masking.
4) after made a edge mask, apply a greyscale mask again with overlay mode at the edge mask. it can preserve more darks.
1) before applying edge detection, raise gamma. it masking more darks, and less brights.
2) apply surface blur after edge detected. it makes surfaces easily mask out with leveling.
3) of cause, leveling is the most important part of edge masking.
4) after made a edge mask, apply a greyscale mask again with overlay mode at the edge mask. it can preserve more darks.
wavelet denoising vs anisotropic denoising
Caution: this post is written by my arbitrary thought. it must be corrected by a professional.
There are 3 types of useful filterings (ignoiring some useless filtering... such as dust&scratch :p).
1) contrast threshold blur (smart blur, surface blur, selective gaussian blur)
2) wavelet denoising (neat image, noise ninja, noise ware)
3) anisotropic denoising (greycstration)
they have different pros and cons.. should be used by what you want to get. I'll point out only cons of 2) and 3) in this post.
A) wavelet denoising
the main problem of wavelet denoising is halo effects around edges. I guess because wavelet denoising hasn't edge detection calculation (need to be explained by a professional. I'm just an amateur) and if you raise the noise levels (thresholds), then edges are smoothed out.
B) anisotropic denoising
there are 2 problems in this filter. one is killing dithering pixels. the gradation of the result image gets "contoured" lines.
the other is about anisotropic noises. if images have noises like lined textures, the filter regard them as edges, and try to make them contour lines. but they are simply ugly :/
Caution: this post is written by my arbitrary thought. it must be corrected by a professional.
There are 3 types of useful filterings (ignoiring some useless filtering... such as dust&scratch :p).
1) contrast threshold blur (smart blur, surface blur, selective gaussian blur)
2) wavelet denoising (neat image, noise ninja, noise ware)
3) anisotropic denoising (greycstration)
they have different pros and cons.. should be used by what you want to get. I'll point out only cons of 2) and 3) in this post.
A) wavelet denoising
the main problem of wavelet denoising is halo effects around edges. I guess because wavelet denoising hasn't edge detection calculation (need to be explained by a professional. I'm just an amateur) and if you raise the noise levels (thresholds), then edges are smoothed out.
B) anisotropic denoising
there are 2 problems in this filter. one is killing dithering pixels. the gradation of the result image gets "contoured" lines.
the other is about anisotropic noises. if images have noises like lined textures, the filter regard them as edges, and try to make them contour lines. but they are simply ugly :/
The important knowledge about scans
Color should be adjusted 2 times
1) while (or after) scanning
2) after (or middle of) filtering
In the 1) process, don't clip out colors, and keep saturation low.
sample color histogram:
the image is already adjusted white/black balance, and also gammas. but levels aren't adjusted tightly because of the following reasons.
1) some filters couldn't work well at 0 and 255
2) filtering changes colors a bit, thus color adjustment can't be done before filtering.
3) but if colors are clipped, no more can be adjusted.
Color adjustment is one of the most hard part of scanning. but in any case, don't clip out colors before get the final best.
if colors aren't clipped out, then someone (or you in the future) can fix them sometime later.
Color should be adjusted 2 times
1) while (or after) scanning
2) after (or middle of) filtering
In the 1) process, don't clip out colors, and keep saturation low.
sample color histogram:
the image is already adjusted white/black balance, and also gammas. but levels aren't adjusted tightly because of the following reasons.
1) some filters couldn't work well at 0 and 255
2) filtering changes colors a bit, thus color adjustment can't be done before filtering.
3) but if colors are clipped, no more can be adjusted.
Color adjustment is one of the most hard part of scanning. but in any case, don't clip out colors before get the final best.
if colors aren't clipped out, then someone (or you in the future) can fix them sometime later.
check "disable color sub sampling" option when you save images as jpeg with Irfanview.midzki said:
Irfanview doesn't have subsampling option. don't use it.
noise types and filterings
scanning_dust
it shows up mainly on bright tones and dark tones. small dusts on bright can be erased easily by ordinary filtering, but other dusts need to be erased by the clone stamp or the patch tool.
paper_texture
it resembles to scanning_dust because they're the issue of physical structure of scanned materials. in other works, they are 3D noises.
this noises on bright tones are eaisly erased by the same way of the above, but I still couldn't get how to kill this noises on darks.
bleed through
get an accurate levels is the first step. then apply strong filter with the color range selection with white, or masking on dark~mid tones.
screening
it exists on mid~dark tones, and separated 2 types of noises.
1) chroma screening noise: just remove CrCb masking brights.
2) luminous screening noise: it mainly exists as white spots on the darks. it's better to use 2 pass filtering (darken+normal) than 1 pass filtering. also don't forget to mask on bright tones.
jpeg_artifacts
poke the man who scanned it.
scanning_dust
it shows up mainly on bright tones and dark tones. small dusts on bright can be erased easily by ordinary filtering, but other dusts need to be erased by the clone stamp or the patch tool.
paper_texture
it resembles to scanning_dust because they're the issue of physical structure of scanned materials. in other works, they are 3D noises.
this noises on bright tones are eaisly erased by the same way of the above, but I still couldn't get how to kill this noises on darks.
bleed through
get an accurate levels is the first step. then apply strong filter with the color range selection with white, or masking on dark~mid tones.
screening
it exists on mid~dark tones, and separated 2 types of noises.
1) chroma screening noise: just remove CrCb masking brights.
2) luminous screening noise: it mainly exists as white spots on the darks. it's better to use 2 pass filtering (darken+normal) than 1 pass filtering. also don't forget to mask on bright tones.
jpeg_artifacts
poke the man who scanned it.
how to change profile: Noiseware
Noiseware is aq similar denoising filter to Neatimage, but it has more advantages to make profiles.
1) when it's run on photoshop action, profiles are recorded in an action file.
2) normal profiles can be edited by notepad.
this is a Noiseware's profile opened with terapad.
<nwp:Values> means noise profiles, and <rdf:_1>~<rdf:11> means high-freq Y noise levels by tones. also <rdf:12>~<rdf:22> means high-freq Cr noise leves. <rdf:34>~<rdf:44> means middle-freq Y noises, e.t.c.
it seems parameter 100 is too strong. I suggest around 50, and very-low should be under 30 in my experience.
Noiseware is aq similar denoising filter to Neatimage, but it has more advantages to make profiles.
1) when it's run on photoshop action, profiles are recorded in an action file.
2) normal profiles can be edited by notepad.
this is a Noiseware's profile opened with terapad.
<nwp:Values> means noise profiles, and <rdf:_1>~<rdf:11> means high-freq Y noise levels by tones. also <rdf:12>~<rdf:22> means high-freq Cr noise leves. <rdf:34>~<rdf:44> means middle-freq Y noises, e.t.c.
it seems parameter 100 is too strong. I suggest around 50, and very-low should be under 30 in my experience.
Windows 7 seems to make filters handle larger images than they do in XP.
With 3GB ram, dual core in windows XP + PS CS3, greyc PS plugin and NI PS Plugin would report low memory for 600dpi A3 sized scans, but in windows 7 they worked well.
I'm not quite clear about the cause, maybe win7 handles memory better, maybe a clean system install, or maybe broken paramenters in my old system... But anyway I give you 1 more reason to switch to windows 7 if you're still wondering ;) .
With 3GB ram, dual core in windows XP + PS CS3, greyc PS plugin and NI PS Plugin would report low memory for 600dpi A3 sized scans, but in windows 7 they worked well.
I'm not quite clear about the cause, maybe win7 handles memory better, maybe a clean system install, or maybe broken paramenters in my old system... But anyway I give you 1 more reason to switch to windows 7 if you're still wondering ;) .
I heard that XP is using 2GB for 1 application, regardless of the maximum of the total memory (3~4GB?). but not sure.
Here is my currently processing. the raw is post #111707 .the result is post #111709.
overview
1) preparation
install Photoshop CS2 or later, Greaycshop, USM2, and Noiseware. then set all noise levels of Noiseware profile 66.
2) resize at 360dpi
you can process 600dpi scan, but the quality of results isn't different than 360dpi so much. I'm resizing scans before any other processing. if scan is FM screening, then just bicubic resizing at 350~360 dpi. if it's AM, add gauss blur 0.5 before resizing.
3) denoise brights
this process reduces noises on only brighter areas using luminance mask.
duplicate layer > Greycstration spoiler > Noiseware spoiler > make a mask at the top layer > apply image (default) > levels 0,250, gamma 0.2 > flatten image.
4) denoise darks
this process reduces noises on only darker areas using luminance mask.
duplicate layer > Noiseware/luminance 0 spoiler > duplicate layer > set layer mode: darken > Noiseware/color 0 spoiler spoiler > merge top 2 layers > Greycstration spoiler at the top layer > make a mask at the top layer > apply image (with inversion) > levels 0,230 > flatten image.
5) sharpen
it's combination of weak Greyc & USM2. because pre-Greyc reduces side effects of USM2.
Greycstration spoiler > USM2 spoiler
6) levels
adjust levels again changing 16bit colors. I can't change mode into 16bit before reached at this step, because of bad allocation ;_;
7) denoise surfaces
this process smooths surfaces using edge mask.
duplicate layer > Noiseware spoiler > make a mask > apply image (default) > gamma 4 spoiler > find edges > surface blur spoiler > levels 150,250
8) manual masking
edge masking of process 7) fails catching some details, thus I have to rewrite it manually.
[edited settings of NW a bit 2010/01/11]
overview
1) preparation
install Photoshop CS2 or later, Greaycshop, USM2, and Noiseware. then set all noise levels of Noiseware profile 66.
2) resize at 360dpi
you can process 600dpi scan, but the quality of results isn't different than 360dpi so much. I'm resizing scans before any other processing. if scan is FM screening, then just bicubic resizing at 350~360 dpi. if it's AM, add gauss blur 0.5 before resizing.
3) denoise brights
this process reduces noises on only brighter areas using luminance mask.
duplicate layer > Greycstration spoiler > Noiseware spoiler > make a mask at the top layer > apply image (default) > levels 0,250, gamma 0.2 > flatten image.
4) denoise darks
this process reduces noises on only darker areas using luminance mask.
duplicate layer > Noiseware/luminance 0 spoiler > duplicate layer > set layer mode: darken > Noiseware/color 0 spoiler spoiler > merge top 2 layers > Greycstration spoiler at the top layer > make a mask at the top layer > apply image (with inversion) > levels 0,230 > flatten image.
5) sharpen
it's combination of weak Greyc & USM2. because pre-Greyc reduces side effects of USM2.
Greycstration spoiler > USM2 spoiler
6) levels
adjust levels again changing 16bit colors. I can't change mode into 16bit before reached at this step, because of bad allocation ;_;
7) denoise surfaces
this process smooths surfaces using edge mask.
duplicate layer > Noiseware spoiler > make a mask > apply image (default) > gamma 4 spoiler > find edges > surface blur spoiler > levels 150,250
8) manual masking
edge masking of process 7) fails catching some details, thus I have to rewrite it manually.
[edited settings of NW a bit 2010/01/11]
It would be easier to see what's happening by putting all images in the same inline, like in forum #5986. That makes it easy to compare each step before/after, to see what each step actually does.
Don't use "Noise -> Median". You'll see some people recommending it, but the results are deceptively poor.
The dark area here is among the hardest type of screening to filter cleanly. The sparse white dots need to be filtered evenly across the surface, or they'll leave ugly little checkered splotches:
The median noise filter, at first glance, seems like it's doing a good job here, removing the white dots for further filtering.
But that's completely wrong. The dots are there for a reason: they affect the brightness of the area. This filter is removing them entirely, which darkens everything and kills luminescence (brightness) detail. The dots aren't there by accident--they need to be filtered, not just removed.
The dark area here is among the hardest type of screening to filter cleanly. The sparse white dots need to be filtered evenly across the surface, or they'll leave ugly little checkered splotches:
The median noise filter, at first glance, seems like it's doing a good job here, removing the white dots for further filtering.
But that's completely wrong. The dots are there for a reason: they affect the brightness of the area. This filter is removing them entirely, which darkens everything and kills luminescence (brightness) detail. The dots aren't there by accident--they need to be filtered, not just removed.
While filtering textured_paper, I leave most noises on darks.
the following state is the difference from http://moe.imouto.org/forum/show/6834
1) in the "4) denoise darks" process, raise the noise levels of "Noiseware/color 0 (darken)" part spoiler, and weakened noise reduction amount at 50%
2) in the "4) denoise darks" process, omit greycstration
3) omit the "5) sharpen" process.
the result is pool #1261.
the following state is the difference from http://moe.imouto.org/forum/show/6834
1) in the "4) denoise darks" process, raise the noise levels of "Noiseware/color 0 (darken)" part spoiler, and weakened noise reduction amount at 50%
2) in the "4) denoise darks" process, omit greycstration
3) omit the "5) sharpen" process.
the result is pool #1261.
The best way of cleaning scans are patching. use it to erase spotted noises on darks.
It was 30sec patching, won't take long time per image if you have a pen tablet (￣▽￣)ノ
It was 30sec patching, won't take long time per image if you have a pen tablet (￣▽￣)ノ
The "healing brush" and "spot healing brush" tools are useful too, especially when you have a lot of small spots to fix. They do the same thing, but you draw a mask instead of a selection.
The patch tool is more precise, particularly when you're aligning patterns (eg. patching over a line), though usually at that point I switch to clone stamp.
The patch tool is more precise, particularly when you're aligning patterns (eg. patching over a line), though usually at that point I switch to clone stamp.
is it possible to export photoshop settings ?midzki said:
Here is my currently processing. the raw is post #111707 .the result is post #111709.
overview
would make a bit easier...
also whats USM2 ?
Here is my action for processing scans
[edit: this post is old and the link is dead]
requirements:
Photoshop CS2
Greycstoration
UnsharpMask2
NoisewareProfessionalPlugin
･it's 600dpi compliant action. but it can be work on 300dpi or higher resolution.
･run batch high-quality (FM) for FM screening thick paper scans, low-quality (AM) for AM screening thin paper scans.
･the results will be saved at the d:\[scans] folder
･move the convert at 16/c process at the top if you have enough RAM.
･the levels process just before the convert at 8/c should be adjusted in each scans.
･make a new batch combinations for middle quality scans. for example, thin FM, thick AM, small dots AM screening, e.t.c.
[edit: this post is old and the link is dead]
requirements:
Photoshop CS2
Greycstoration
UnsharpMask2
NoisewareProfessionalPlugin
･it's 600dpi compliant action. but it can be work on 300dpi or higher resolution.
･run batch high-quality (FM) for FM screening thick paper scans, low-quality (AM) for AM screening thin paper scans.
･the results will be saved at the d:\[scans] folder
･move the convert at 16/c process at the top if you have enough RAM.
･the levels process just before the convert at 8/c should be adjusted in each scans.
･make a new batch combinations for middle quality scans. for example, thin FM, thick AM, small dots AM screening, e.t.c.
Maybe you should provide the working dpi?
thx, will try to get this under cs4-64 to work
as i plan to scan stuff soon, and have almost no experience in cleaning (besides basic photoshop filters)
good, but neither usm2 nor grey work on 64bit version
32bit works fine though
worked kinda good on a 1200 dpi scan
as i plan to scan stuff soon, and have almost no experience in cleaning (besides basic photoshop filters)
good, but neither usm2 nor grey work on 64bit version
32bit works fine though
worked kinda good on a 1200 dpi scan
midzki
Original ai vector format: 379kb :o
32bit png: 916kb
24bit jpeg 90% (acceptable quality): 1459kb
24bit jpeg 100% (worst setting): 3808kb
24bit jpegged png (recompressed jpeg90%): 3283kb
ps: at a game cg (post #61237)
png original: 1845kb
jpeg 90: 450kb
jpeg 100: 2055kb
jpegged png: 1856kb
jpeg_artifacts is clearly visible on jpeg 90% images without enlarged view.
don't use this quality. at least, use 95% without down sampling.
in game_cg, file size of 3×jpeg95 ≒ lossless png
caution: even if there are no visible jpeg_artifacts, jpegging always decrease color depth (desaturate vivid red, blocking darks, e.t.c.).
this color shift is a serious matter in processing of scans.