Ad
Ad
Ad
Pages: « 1 2 [3]   Bottom of Page
Print
Author Topic: Noise removal?  (Read 18085 times)
bjanes
Sr. Member
****
Offline Offline

Posts: 2714



« Reply #40 on: June 25, 2008, 10:05:49 AM »
ReplyReply

Quote
Bill,

do you mind uploading the raw of this demo?

Thanks
[{POST_SNAPBACK}][/a]

Not at all. Please let us know about your findings.

Bill


[a href=\"https://download.yousendit.com/2B85914E523CA118]https://download.yousendit.com/2B85914E523CA118[/url]
Logged
gmitchel
Full Member
***
Offline Offline

Posts: 151


WWW
« Reply #41 on: June 25, 2008, 04:06:26 PM »
ReplyReply

I am troubled by the sample images here.

I understand the use of greatly magnified examples to distinctly point out noise to others.

However, if you are going to compare noise reduction workflows or the results of differnet noise reduction tools, the examples should be at print size (or display size for Web images).

Grossly magnified samples of noise reduction are just as misleading as grossly magnified samples of sharpening.

I have many prints where you would need a loupe to see the noise, yet at 100% on the monitor, noise is very obvious. Especially with images from cameras like the Canon 1Ds MkII, 1Ds MkIII, etc.

If the goal is to understand how noise reduction and capture sharpening interact, the magnified samples on this thread are really not all that helpful.

Cheers,

Mitch
Logged
Mark D Segal
Contributor
Sr. Member
*
Offline Offline

Posts: 6767


WWW
« Reply #42 on: June 25, 2008, 04:37:26 PM »
ReplyReply

Quote
I am troubled by the sample images here.

IIf the goal is to understand how noise reduction and capture sharpening interact, the magnified samples on this thread are really not all that helpful.

Cheers,

Mitch
[a href=\"index.php?act=findpost&pid=203671\"][{POST_SNAPBACK}][/a]

This I don't agree with. I've found that I need 100% magnification to really see what's going on with my 1Ds3 files. That doesn't mean the display view is what will show in a letter size, or even 13*19 inch print - but that's not the point. Who knows, one day I may wish to repurpose a file for yet much larger sizes, and then what's really there could show in a print as it does magnified on the display. Also, for analytical purposes it is necessary to magnify the image to the extent needed to see what the file consists of. The only risk here is confusing pixellation with sharpening artifacts and noise, but I assume those of us doing this work are aware of that and know how to tell the difference.
Logged

Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....." http://www.luminous-landscape.com/reviews/film/scanning_workflows_with_silverfast_8.shtml
bjanes
Sr. Member
****
Offline Offline

Posts: 2714



« Reply #43 on: June 25, 2008, 05:33:25 PM »
ReplyReply

Quote
I am troubled by the sample images here.

I understand the use of greatly magnified examples to distinctly point out noise to others.

However, if you are going to compare noise reduction workflows or the results of differnet noise reduction tools, the examples should be at print size (or display size for Web images).

Grossly magnified samples of noise reduction are just as misleading as grossly magnified samples of sharpening.

I have many prints where you would need a loupe to see the noise, yet at 100% on the monitor, noise is very obvious. Especially with images from cameras like the Canon 1Ds MkII, 1Ds MkIII, etc.

If the goal is to understand how noise reduction and capture sharpening interact, the magnified samples on this thread are really not all that helpful.

Cheers,

Mitch
[a href=\"index.php?act=findpost&pid=203671\"][{POST_SNAPBACK}][/a]


If I had a 360 ppi monitor, then previewing on screen at the actual print size would be a good idea. As is, I need a higher magnification.

Bill
Logged
gmitchel
Full Member
***
Offline Offline

Posts: 151


WWW
« Reply #44 on: June 25, 2008, 08:24:27 PM »
ReplyReply

Quote
This I don't agree with. I've found that I need 100% magnification to really see what's going on with my 1Ds3 files. That doesn't mean the display view is what will show in a letter size, or even 13*19 inch print - but that's not the point. Who knows, one day I may wish to repurpose a file for yet much larger sizes, and then what's really there could show in a print as it does magnified on the display. Also, for analytical purposes it is necessary to magnify the image to the extent needed to see what the file consists of. The only risk here is confusing pixellation with sharpening artifacts and noise, but I assume those of us doing this work are aware of that and know how to tell the difference.
[{POST_SNAPBACK}][/a]

Noise reduction is antagonistic towards sharpening and image details. We should, therefore, only reduce noise to the point that it is no longer a distracting visual artifact.

As George Bowen at Imagenomic told me today, "Pixel peeping at anything other than intended output size is a waste of time, especially in the noise reduction industry." Here's a .PDF from Noiseware on workflow:

[a href=\"http://www.imagenomic.com/downloads/NWNRWorkflow.pdf]http://www.imagenomic.com/downloads/NWNRWorkflow.pdf[/url]

As the PDF indicates, Print Size and 100% Magnification give two very different perceptions about noise. "You can drive yourself crazy trying to remove all noise from an image at higher screen magnifications, when you will achieve much better results at a view that corresponds to your output." [Pgs. 7-8]

George and I were chatting via e-mail because I am going to extend my reviews of noise reduction software and sharpening software.

There is no "one setting fits all resampling" setting for noise reduction software. You will almost certainly reduce noise too much by using higher magnification as your guide and obliterate fine details that could be preserved. "Be careful to not fall into the rut that some advocate of doing noise reduction at high magnification and then touting that the downward resized image will look even better. In all but minimal cases, the smaller image will lose a lot of detail using this method." [Pg. 8]

As with sharpening, your monitor is not a good guide for final sharpening settings. Prints can show noise that a monitor cannot reproduce. You need a hard proof.

You should always apply noise reduction on a layer because you might need to adjust the settings later. Tonal adjustments will affect the visibility of luminosity noise. Color adjustments will affect the visibility of color noise. You can find that your noise reduction settings were not optimal after you finish adjustments to color, tone, and sharpening. It helps to be able to go back and substitute other settings.

I'll close with this: "The key to attaining an optimum level of noise reduction is determined by your output type and size. There is no generic "one size fits all" approach that will allow optimum reduction for all of the various outputs that are available to an image. A small web viewed image will not need the same level of noise reduction as an 8x10 print, and the 8x10 print will require different reduction methods than will a 20x30 print. Furthermore, print output type, whether continuous tone or inkjet based, and paper types will play important roles." [Pg. 6]

The approach you propose is more efficient. Run noise reduction once to make a master file and then use that to resample into several different output sizes. As the "Noise Reduction Workflow Tutorial" makes very clear, efficient is not optimal in this case.

I invite you to ask Kent Christiansen at PictureCode (Noise Ninja) or Vlad at NeatImage. They'll likely give you the same advice.

Cheers,

Mitch
« Last Edit: June 25, 2008, 08:28:53 PM by gmitchel » Logged
Gordon Buck
Sr. Member
****
Offline Offline

Posts: 409



WWW
« Reply #45 on: June 25, 2008, 08:53:19 PM »
ReplyReply

When my digital camera had only  3/4 MP, I looked at the images "full screen".  Is it permissible to crop my many megapixels such that the remaining cropped image is full screen?
Logged

gmitchel
Full Member
***
Offline Offline

Posts: 151


WWW
« Reply #46 on: June 25, 2008, 09:30:05 PM »
ReplyReply

Quote
When my digital camera had only 3/4 MP, I looked at the images "full screen". Is it permissible to crop my many megapixels such that the remaining cropped image is full screen?
[a href=\"index.php?act=findpost&pid=203712\"][{POST_SNAPBACK}][/a]

At 1024x768 screen resolution, your 3/4 MP camera image was 1:1 in terms of screen pixels and image pixels. But your print resolution was still probably something like 180 ppi, which would still give you approximately a 200% zoom relative to print size.

"Learn to view your noise reduction in a manner that coincides with your intended output. Trying to remove all noise in 400% zooms will cause you to age prematurely." [Pg. 74] You were using a 200% zoom, not quite so bad. The added stress will just shorten your lifespan. LOL.

I have used a 200% zoom or even a 400% zoom to get an idea about whether noise is likely to become visible after corrections to color and tone and after sharpening. With sharpening, I pixel peep, because I can translate optimal sharpening into sharpening halo contour width. I do not use those zooms to determine my noise reduction settings.

Have you ever seen a RAW file from a Canon 1Ds MkII or MkIII at 100% zoom? It is like looking at a poster size print with your face a few inches away. Go to 200% or 400% and yikes, it's like looking at a billboard a few feet away! Tiny imperfections that the eye would almost cerftainly miss at print size stand out, and the inclination is to reduce those artifacts. That leads to smearing and loss of detail from too much noise reduction.

Using too much magnification is guaranteed to scare you, even with a camera like the 1Ds MkII or MkIII. You can see individual pixel-level abnormalities on the monitor that would require a loupe to see on a print and invisible to the eye at normal viewing distance.

Cheers,

Mitch
« Last Edit: June 25, 2008, 09:34:18 PM by gmitchel » Logged
Mark D Segal
Contributor
Sr. Member
*
Offline Offline

Posts: 6767


WWW
« Reply #47 on: June 25, 2008, 09:58:56 PM »
ReplyReply

Quote
As George Bowen at Imagenomic told me today, "Pixel peeping at anything other than intended output size is a waste of time, especially in the noise reduction industry." Here's a .PDF from Noiseware on workflow:


Cheers,

Mitch
[a href=\"index.php?act=findpost&pid=203705\"][{POST_SNAPBACK}][/a]

Self-serving advice, and the rest of his incantation ignores the fact that people re-purpose images. He may be good at writing software, but I'm not impressed with this approach to using it. At 100% I can see both noise and image detail, especially with everything I do on separate layers, so I don't have any problem assessing the trade-off between noise reduction and image detail. The main thing in Photoshop is to use screen magnifications of 25%, 50%, 75% or 100% because the image is aliased at intermediate values.

I should add to this, that the Noiseware preview shows the image at 100% by default.
« Last Edit: June 25, 2008, 10:02:38 PM by MarkDS » Logged

Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....." http://www.luminous-landscape.com/reviews/film/scanning_workflows_with_silverfast_8.shtml
gmitchel
Full Member
***
Offline Offline

Posts: 151


WWW
« Reply #48 on: June 25, 2008, 11:15:53 PM »
ReplyReply

Quote
Self-serving advice, and the rest of his incantation ignores the fact that people re-purpose images. He may be good at writing software, but I'm not impressed with this approach to using it. At 100% I can see both noise and image detail, especially with everything I do on separate layers, so I don't have any problem assessing the trade-off between noise reduction and image detail. The main thing in Photoshop is to use screen magnifications of 25%, 50%, 75% or 100% because the image is aliased at intermediate values.

I should add to this, that the Noiseware preview shows the image at 100% by default.
[a href=\"index.php?act=findpost&pid=203723\"][{POST_SNAPBACK}][/a]

LOL. Since Noiseware has no idea what size your image will be printed and at what resolution, how could it display the image at a magnification close to print size?!

Why would you gratuitously insult George by referring to his comment as self-serving?! What evidence do you have for such a gratuitous and outrageous personal attack? I don't know that George wrote the Imagenomics PDF file. It says it is authored by the Imagenomics Team.

Who wrote the PDF is irrelevant. Stop the ad hominem and deal with the logic. That same tutorial could have been written for Noise Ninja, NeatImage, etc.

If you want to claim superior judgment to the people who engineer noise reduction software, share your credentials. I doubt you have tested as many images, from as many cameras and scanners, under as many conditions as any of the people who engineer noise reduction software. Maybe you have?!

You have not adequately addressed the fact that noise is over-emphasized with magnified zooms. It is much more apparent than it is at print size. Do you deny this?!

If not, then how can reducing an over-emphasized representation of the visual impact of the noise lead to optimal noise reduction?!

You seem to assume that the relationship between noise and image details is constant at all output sizes and output media. That simply is not the case at all. You can easily prove this to yourself by taking any noisy photo and reducing it or enlarging it.

Since it is not the case that the relationship between visible noise artifacts and image detail is fixed, you can only get to optimal noise reduction for a print (which is just enough to make the noise non-apparent to the eye under expected viewing conditions and no more) by hard proofing. You can come close with a monitor by using a zoom that is near to the intended output size (near being a reference to aliasing).

If your zoom overemphasizes noise and you react to that, do you then pull back on the settings to compensate for overemphasis from your magnified view to avoid too much noise reduction?!?! I'll bet you do not. And that, Mark, means you almost certainly lose detail that could be preserved with better technique.

Cheers,

Mitch
« Last Edit: June 25, 2008, 11:26:02 PM by gmitchel » Logged
Mark D Segal
Contributor
Sr. Member
*
Offline Offline

Posts: 6767


WWW
« Reply #49 on: June 26, 2008, 07:05:02 AM »
ReplyReply

Quote
LOL. Since Noiseware has no idea what size your image will be printed and at what resolution, how could it display the image at a magnification close to print size?!

Why would you gratuitously insult George by referring to his comment as self-serving?! What evidence do you have for such a gratuitous and outrageous personal attack? I don't know that George wrote the Imagenomics PDF file. It says it is authored by the Imagenomics Team.

Who wrote the PDF is irrelevant. Stop the ad hominem and deal with the logic. That same tutorial could have been written for Noise Ninja, NeatImage, etc.

If you want to claim superior judgment to the people who engineer noise reduction software, share your credentials. I doubt you have tested as many images, from as many cameras and scanners, under as many conditions as any of the people who engineer noise reduction software. Maybe you have?!

You have not adequately addressed the fact that noise is over-emphasized with magnified zooms. It is much more apparent than it is at print size. Do you deny this?!

If not, then how can reducing an over-emphasized representation of the visual impact of the noise lead to optimal noise reduction?!

You seem to assume that the relationship between noise and image details is constant at all output sizes and output media. That simply is not the case at all. You can easily prove this to yourself by taking any noisy photo and reducing it or enlarging it.

Since it is not the case that the relationship between visible noise artifacts and image detail is fixed, you can only get to optimal noise reduction for a print (which is just enough to make the noise non-apparent to the eye under expected viewing conditions and no more) by hard proofing. You can come close with a monitor by using a zoom that is near to the intended output size (near being a reference to aliasing).

If your zoom overemphasizes noise and you react to that, do you then pull back on the settings to compensate for overemphasis from your magnified view to avoid too much noise reduction?!?! I'll bet you do not. And that, Mark, means you almost certainly lose detail that could be preserved with better technique.

Cheers,

Mitch
[a href=\"index.php?act=findpost&pid=203733\"][{POST_SNAPBACK}][/a]

Calm down. I'm not personally attacking anyone. I'm criticizing a workflow recommendation as favoring the apparent quality of the product versus that of the image.

Noiseware is an excellent product and I have every respect for it's author's software engineering skills, as I said, but when someone tells me only to look for artifactss at a magnification where they don't appear, but they may still exist, what am I supposed to take away from that, when I really don't think it's good technical advice? You still haven't dealt with my point that people re-purpose images. Once you've flattened your noise reduction layer, saved and quit, you've cooked your goose.

The logic I'm relating here is very simple - and very logical - zoom the image to the magnification needed to see what's really there. If you read what I said, I did not say that Noiseware uses a magnification reflecting print size. I said its default magnification is 100% - in the usual meaning of display magnification - i.e. display pixels, not print.

If you read what I said, you'll see that I have already acknowledged that 100% magnification will exaggerate the extent of VISIBLE noise relative to letter size and perhaps even 13*19 inch prints I would obtain from my 1Ds3, but that's not the central issue I'm getting at. I'm talking about the magnification needed to see what's really embedded in the image regardless of print size. And yes, I agree, what's VISIBLE depands on size, but that's not the point.  

I know my display, the software and print conditions well enough that I don't need to hardproof to see what I'm getting. I can optimise between noise and sharpening at 50% or 100%, then return to 25%~50% depending on print size and that gives me a reliable indicator of what VISIBLE EFFECT to expect out of the printer. This comes with some experience and lack of image detail hasn't been a complaint from professional peer reviewers.

You are making all this seem more ephemeral and difficult than it really is. Furthermore, with appropriate Noiseware settings, selective noise reduction, separate layers and the use of opacity controls, there's no reason to overcompensate either noise reduction or sharpening whether at 100% magnification or lower.

Take it easy Mitch. This little tempest in a teapot started in support of an analytical approach presented by Bill Janes which I thought quite useful and you didn't. So be it and AMEN.
Logged

Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....." http://www.luminous-landscape.com/reviews/film/scanning_workflows_with_silverfast_8.shtml
bjanes
Sr. Member
****
Offline Offline

Posts: 2714



« Reply #50 on: June 26, 2008, 08:49:27 AM »
ReplyReply

Quote
Noise reduction is antagonistic towards sharpening and image details. We should, therefore, only reduce noise to the point that it is no longer a distracting visual artifact.

As with sharpening, your monitor is not a good guide for final sharpening settings. Prints can show noise that a monitor cannot reproduce. You need a hard proof.
[a href=\"index.php?act=findpost&pid=203705\"][{POST_SNAPBACK}][/a]

Your points about pixel peeping are well taken. However, the high magnification views that I posted were not to preview the appearance of noise at a given print size (in which case your criticism is valid), but rather to demonstrate what degree of noise is present in the image.


Quote
I invite you to ask Kent Christiansen at PictureCode (Noise Ninja) or Vlad at NeatImage. They'll likely give you the same advice.

[a href=\"index.php?act=findpost&pid=203705\"][{POST_SNAPBACK}][/a]

Jim Christian is the force behind Noise Ninja
Logged
gmitchel
Full Member
***
Offline Offline

Posts: 151


WWW
« Reply #51 on: June 26, 2008, 10:44:16 AM »
ReplyReply

> Self-serving advice, and the rest of his incantation

That is not exactly dispassionate, Mark. That's very charged language. You could have disagreed about the workflow recommendation without the gratuitous attack. Or, perhaps you think "self-serving" and linking someone to witchcraft or alchemy is benign or neutral language?

I am and remain perfectly calm. I can disagree with you without getting animated. LOL.

> talking about the magnification needed to see what's really embedded in the
> image regardless of print size. And yes, I agree, what's VISIBLE depands on
> size, but that's not the point.

Who cares about noise embedded in the image regardless of print size?!?! What's visible is entirely the point.

> when someone tells me only to look for artifactss at a magnification where they
> don't appear, but they may still exist, what am I supposed to take away from
> that

That you need to use some common sense. As a digital photographer, what matters is the visible effect of noise on our photographs. If your eye is not distracted by the noise that's present in a photograph, where's the problem?! Leave it alone.

What does it matter, if you intend to print at 8x12 and the print has no visible noise *BUT* a 400% magnification with a loupe shows evidence of noise? How does that diminsh the print in any way. Do you print with the intended viewing distance being the eyepiece end of a loupe?! If so, then go for the extreme magnification.

Noise reduction comes at a high cost. You only want to apply enough to make noise non-apparent (or less apparent in the case of extreme noise). Any more than that and you sacrifice image detail.

What you want to do is pretend that there is one setting for noise reduction that is suitable for a master file that can be reproduced at different sizes and resolutions. That's simply not the case. Ignoring that and targeting the maximum size you might ever intend to print or trying to remove every vestige of noise will certainly erode detail for smaller prints that could have been retained by targeting the noise reduction for the intended output.


> I know my display, the software and print conditions well enough that I don't
> need to hardproof to see what I'm getting.

LOL. You had the nerve to apply "self-serving" to the comments of another!?

Rubbish. Monitors emit light. Prints reflect it. Not all noise visible on a print is visible on a monitor and vice versa. You can come close with a soft proof.


> However, the high magnification views that I posted were not to preview the
> appearance of noise at a given print size (in which case your criticism is valid),
> but rather to demonstrate what degree of noise is present in the image.

I understand all of that. You have one perspective. Mark, however, is advocating a different workflow. One that assumes a constant relationship between the appearance of noise and detail.

The comparison of how much noise exists at high magnification is still misleading, however. Since that is not the size of the intended output (maybe not even the intended resolution and media), differences in noise might become much less significant (or even insignificant) at more appropriate magnification.

It doesn't matter if the comparison results in significant differences in noise at high magnification, unless that high magnification is what you intend for the photo. It only matters if significant differences exist in the final output.

If you want to compare what happens to a photo as a result of noise reduction and capture sharpening, hold everything else constant through output and then view that output. If a Web image is the final output, look at that. You don't need magnification to see visible artifacts from noise that catches the eye. If it's not apparent to the eye, it's not worthy of comparison.

What you're doing is like the audiophile who sees noise on an oscilloscope. Who cares if you can see audio noise as a waveform?! Can you hear it, is the relevant question when you're buying audio gear.

If you cannot see a significant difference in your output, who cares if you can see it with a loupe (or the same thing, high magnification on a monitor)? Just enjoy your print and don't sweat what cannot be seen under proper viewing conditions.

Cheers,

MItch
« Last Edit: June 26, 2008, 11:00:20 AM by gmitchel » Logged
gmitchel
Full Member
***
Offline Offline

Posts: 151


WWW
« Reply #52 on: June 26, 2008, 10:49:06 AM »
ReplyReply

Quote
Jim Christian is the force behind Noise Ninja
[a href=\"index.php?act=findpost&pid=203791\"][{POST_SNAPBACK}][/a]

LOL. Yes. I've spent too much time on DPR lately. Kent Christiansen is a very active member there.

My apology to Jim.

Thanks for the nudge.

Cheers,

Mitch
« Last Edit: June 26, 2008, 10:57:47 AM by gmitchel » Logged
Mark D Segal
Contributor
Sr. Member
*
Offline Offline

Posts: 6767


WWW
« Reply #53 on: June 26, 2008, 01:23:39 PM »
ReplyReply

Quote
> Self-serving advice, and the rest of his incantation

That is not exactly dispassionate, Mark. That's very charged language. You could have disagreed about the workflow recommendation without the gratuitous attack. Or, perhaps you think "self-serving" and linking someone to witchcraft or alchemy is benign or neutral language?

I am and remain perfectly calm. I can disagree with you without getting animated. LOL.


MItch
[a href=\"index.php?act=findpost&pid=203807\"][{POST_SNAPBACK}][/a]

No personal attack against anyone was intended; if it came accross that way I'm sorry about that; time to move on.

Turning to the substance, I think all relevant angles on this subject have been aired, and readers can test for themselves what workflow they prefer. I have nothing more to add to it.
Logged

Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....." http://www.luminous-landscape.com/reviews/film/scanning_workflows_with_silverfast_8.shtml
Pages: « 1 2 [3]   Top of Page
Print
Jump to:  

Ad
Ad
Ad