| Author | Thread |
|
|
06/15/2004 10:15:42 AM · #26 |
oops! sorry guys i didnt mean to start anything this involved, it was more of an equation of neatimage to turning your photos into melted butter. i agree that some images look pretty cool through NI, but lets not paint with too broad a brush, here. do what needs to be done. thats it.
Message edited by author 2004-06-15 10:17:52.
|
|
|
|
06/15/2004 10:20:25 AM · #27 |
For images with a lot of noise, especially after several steps of editing, I sometimes use a combination of camera profile, area sampling and auto fine tune in Neat Image.
Do most people use NI at the beginning or at the end of their workflow?
|
|
|
|
06/15/2004 10:41:52 AM · #28 |
Originally posted by BobsterLobster:
Originally posted by Gordon:
Originally posted by BobsterLobster:
It's almost impossible to run NeatImage properly on anything but a full sized image. NeatImage needs a decently sized area of photo from which to judge what is noise and what is detail. |
For what its worth, this is the least effective way to use neatimage. Better is to characterise the noise profile of your camera at particular ISOs and JPEG or RAW settings and just apply it that way. Then you don't need to fuss around trying to sample flat areas of an image. This is all well explained in the documentation. |
I disagree. It may be quicker, but noise reduction will always by definition be carried out better by defining flat areas. Unless you have a photo that has had NO levels or curves adjustments, you will need to define a flat area after these adjustments. |
The noise is a function of the image sensor and settings you used (and to a certain extent temperature and shutter speed)
There is _nothing_ that is image specific about noise, hence profiling the same camera each time on each new image just doesn't make sense and will almost always lead to a suboptimal profile being created.
It is simple to create a set of noise profiles for you camera and use them in each case - in fact the neat image tool is set up to be used this way in anything beyond the most basic operation. The profiling operation uses large, flat areas of known tonality to build up a profile across all tonal ranges in an image - apart from exceptionally unusual images, you will get a much better noise profile from a target than you can build from a particular real image.
Message edited by author 2004-06-15 10:44:37.
|
|
|
|
06/15/2004 10:51:29 AM · #29 |
I think you'll find that the following forum entry by someone who works for NeatImage will back up what I'm saying... namely that it is better to sample each photo rather than use a profile.
//www.neatimage.net/forum/viewtopic.php?t=789 |
|
|
|
06/15/2004 10:56:56 AM · #30 |
Originally posted by BobsterLobster:
I think you'll find that the following forum entry by someone who works for NeatImage will back up what I'm saying... namely that it is better to sample each photo rather than use a profile.
//www.neatimage.net/forum/viewtopic.php?t=789 |
Funnily enough, I'd say it backs up what I'm saying really ;) Also not that it refers to using downloaded profiles. Certainly they'll give worse results than profiles based on the specific camera that you are using. It is always best to calibrate a noise profile to the actual camera, not to the generic class of cameras that it belongs to. So I'd agree, you are better off making a profile from an image taken with the camera - but a specific target, using that camera will give better results than any given random image you want to try and create a profile from.
If your images typically have large, flat areas in all of the dark, mid and light tonal ranges, then probably a profile based on that particular image will be slightly more accurate, as it takes in to account the particular sensor noise (and temperature and shhutter speed) at the instant the shot was taken. That's the only real variant in the noise on the sensor - the image has nothing to do with the noise in it. Often people make the mistake of for example shooting a blue sky to make a noise profile too - again, this only really lets you sample one part of the tonal range/ noise profile - it ignores almost all of the tonal range.
I don't typically take shots that look like test cards though, so large flat areas are often only in one or two of the regions required and carefully generated profiles will always have the full gamut of tonality evaluated, compared to an image based profile. Some level of 'auto tuning' after profile selection can dial in to specific temperature changes between the test target and profile and give a best of both worlds approach, without needing large flat tonally spread regions.
Message edited by author 2004-06-15 11:02:26.
|
|
|
|
06/15/2004 11:01:29 AM · #31 |
Well, perhaps this forum entry and reply by someone who works at NeatImage will definitively put the stamp of approval on what I'm saying...
//www.neatimage.net/forum/viewtopic.php?t=298 |
|
|
|
06/15/2004 11:03:05 AM · #32 |
| Phew, good grief! Do I win yet?! ;-) |
|
|
|
06/15/2004 11:11:08 AM · #33 |
Originally posted by BobsterLobster:
Phew, good grief! Do I win yet?! ;-) |
That would assume you'd ever read what I was saying and respond to it :)
As I've said half a dozen times now - if you take images with large, uniform regions at all the tonal ranges you want to remove noise at, then creating a profile from that image will give the best consideration of the ambient temperature when you took the shot. That's the only differential between a noise profile produced for the same camera, with the same settings. I almost never take shots that have large, uniform regions at ten or more tonal ranges, as a result, properly calibrated images (using the camera that took the image to be considered, not ones downloaded from the internet) will give consistently better results for real world images. Auto fine tuning can be used to tweak that target t o suit for areas in the image that can be auto tuned (i.e., large enough uniform regions)
The problem is people do profiling for particular areas they care about, and ignore the rest of the image (because it doesn't have large tonal blocks at say shadows or highlights or mid tones) so only profile a part of the tonal range. Using a calibration avoids this half profile approach - which is why all the NITeam recommend profiles + auto fine tune by the way...
The various posts from people who work for neat image, that you've linked to, and others on that site all say exactly the same thing that I've been saying.
Noise is a function of the camera and the settings. It has nothing to do with the particular image, other than temperature related differences with the profile.
Message edited by author 2004-06-15 11:13:46.
|
|
|
|
06/15/2004 11:21:32 AM · #34 |
I don't think we're on the same planet here... the administrator of the forum clearly says that if it is at all possible it's better to adjust NeatImage per individual image.
This is in direct contrast to what you are saying, that NeatImage is better used with a profile, rather than tuning to each individual image.
Sometimes, you have to use the profile if there are no flat areas... this is not that often in most people's photography.
You seem to be backtracking by saying that NeatImage can be callibrated for certain types of photos... this seems way too complicated and is hiding the fact that NeatImage themselves recommend callibrating to individual images wherever possible.
I hear what you're saying about noise staying constant per digital cameras, but I have no proof of this, and I've found that my experience with using noise reduction says otherwise.
Bob |
|
|
|
06/15/2004 11:23:41 AM · #35 |
| Why do you need 10 or more tonal ranges in your flat areas? Have I missed something? |
|
|
|
06/15/2004 11:28:49 AM · #36 |
From the user manual:
note the following phrase, and NOTHING about 10 tonal ranges or more...
from the MOST desirable to the LEAST desirable from the standpoint of accuracy
List of profiling techniques, from the most desirable to the least desirable from the standpoint of accuracy:
Select a 100x100+ uniform feature-less patch in your working image and build a rough profile; (optionally) fine-tune the profile manually (use more feature-less patches, the larger in size the better) or using auto fine-tuning;
Do (1) with another image from your camera/scanner shot/scanned in similar conditions; (optionally) additionally fine-tune it with your working image (equalizer values shouldn't change much when images are shot in similar conditions);
Do (1) with another image from another camera/scanner of the same model shot/scanned in similar conditions; (optionally) additionally fine-tune it with your working image (equalizer values shouldn't change much when images are shot in similar conditions);
Get a ready-made profile made with a similar image from another camera/scanner of the same model; (optionally) additionally fine-tune it with your working image (equalizer values shouldn't change much when images are shot in similar conditions);
Do (1) with a smaller (60x60+) patch from your working image;
Cut out a 59x59- uniform feature-less patch and (preferably) seamlessly clone it in an image editor to produce a 60x60+ area; do (1) with the resulting larger area.
Up-size your image (using your favorite method) in an image editor; do (1) with it; process the upsized image in NI (do not process original image with such a profile); down-size the result in the image editor.
|
|
|
|
06/15/2004 11:30:40 AM · #37 |
Originally posted by BobsterLobster:
I don't think we're on the same planet here... the administrator of the forum clearly says that if it is at all possible it's better to adjust NeatImage per individual image.
This is in direct contrast to what you are saying, that NeatImage is better used with a profile, rather than tuning to each individual image.
Sometimes, you have to use the profile if there are no flat areas... this is not that often in most people's photography.
You seem to be backtracking by saying that NeatImage can be callibrated for certain types of photos... this seems way too complicated and is hiding the fact that NeatImage themselves recommend callibrating to individual images wherever possible.
I hear what you're saying about noise staying constant per digital cameras, but I have no proof of this, and I've found that my experience with using noise reduction says otherwise.
Bob |
I'm not backtracking.
I'm also agreeing with what the neat image people are saying.
Noise is a function of the sensor and the ambient conditions the sensor is in. I have plenty of proof of this, including technical documents on my desk so it is something I understand a fair bit about.
The neat image profiles (either from a calibration target, or from a given suitable image) profile the noise at a variety of tonal ranges (from dark, through to light, in about 10 bins)
A calibration target, by design, has large, flat, featureless regions at each of these tonal ranges. By shooting the target at the various camera settings that influence noise (ISO, white balance & compression settings and that's about it for meaningful differences) You create a set of profiles, which have equal representation in each of those bins, giving you the optmial spread for a particular set of camera parameters.
Those parameters drift slightly due to environment factors, such as temperature, which are not currently captured in EXIF data (but could potentially be with the correct sensors on the camera)
The majority if real world images do not have suitable areas for profile generation at all tonal ranges. The majority of image based profiles are thus a compromise. It does represent the best potential basis for a profile, if you have suitable large regions at all 10 of the bin tonal ranges. This is typically unrealistic. The various forum posts from neat image representatives suggest using a calibration target/ profile based approach to get the best spread of optimisation but then tweak that using the image (for the specific tonal regions that are represented in the image)
If you have images with all the large flat regions at all tonal ranges, then yes that is the best option, due to the external factors. However, this is usually an entirely unrealistic situation. For the majority of use, using profiles (built on the camera) with auto fine tuning can give the best result, combinging the spread of a custom calibration target and the up to the minute ambient temperature noise profile from the real image.
|
|
|
|
06/15/2004 11:32:33 AM · #38 |
Originally posted by BobsterLobster:
Why do you need 10 or more tonal ranges in your flat areas? Have I missed something? |
Just look at the profile generated. It is adjusted in a variety of tonal ranges.
Also look at the recommended calibration target.
Calibration target
Also consider what you know about noise - it is not consistant across the tonal range of an image.
You want a profile (no matter how you generate it) that has evaluated the noise in all the tonal regions you want to operate upon. This is how neat image actually works.
NB: I looked at a screen shot. 9 bins, not 10.
Message edited by author 2004-06-15 11:34:29.
|
|
|
|
06/15/2004 11:34:15 AM · #39 |
Originally posted by Gordon:
I'm not backtracking.
I'm also agreeing with what the neat image people are saying.
Noise is a function of the sensor and the ambient conditions the sensor is in. I have plenty of proof of this, including technical documents on my desk so it is something I understand a fair bit about.
The neat image profiles (either from a calibration target, or from a given suitable image) profile the noise at a variety of tonal ranges (from dark, through to light, in about 10 bins)
A calibration target, by design, has large, flat, featureless regions at each of these tonal ranges. By shooting the target at the various camera settings that influence noise (ISO, white balance & compression settings and that's about it for meaningful differences) You create a set of profiles, which have equal representation in each of those bins, giving you the optmial spread for a particular set of camera parameters.
Those parameters drift slightly due to environment factors, such as temperature, which are not currently captured in EXIF data (but could potentially be with the correct sensors on the camera)
The majority if real world images do not have suitable areas for profile generation at all tonal ranges. The majority of image based profiles are thus a compromise. It does represent the best potential basis for a profile, if you have suitable large regions at all 10 of the bin tonal ranges. This is typically unrealistic. The various forum posts from neat image representatives suggest using a calibration target/ profile based approach to get the best spread of optimisation but then tweak that using the image (for the specific tonal regions that are represented in the image)
If you have images with all the large flat regions at all tonal ranges, then yes that is the best option, due to the external factors. However, this is usually an entirely unrealistic situation. For the majority of use, using profiles (built on the camera) with auto fine tuning can give the best result, combinging the spread of a custom calibration target and the up to the minute ambient temperature noise profile from the real image. |
But that is not what the manual says.
The manual says that THE BEST AND MOST PREFERABLE way to proceed with an image is to do it with a 100*100 flat area.
I never read the manual, but found that through trial and error, this is the best way that works for me.
Arguing any further is going to be futile, I feel.
Let's agree to disagree. |
|
|
|
06/15/2004 11:35:58 AM · #40 |
nm
Message edited by author 2004-06-15 11:37:29.
|
|
|
|
06/15/2004 11:37:00 AM · #41 |
Originally posted by Gordon:
Originally posted by BobsterLobster:
Why do you need 10 or more tonal ranges in your flat areas? Have I missed something? |
Just look at the profile generated. It is adjusted in a variety of tonal ranges.
Also look at the recommended calibration target.
Calibration target
I can see the point of using 10 tonal ranges in a profile, but where does it say that you need 9/10 or more tonal ranges when profiling a flat area in an individual photo?
Also consider what you know about noise - it is not consistant across the tonal range of an image.
You want a profile (no matter how you generate it) that has evaluated the noise in all the tonal regions you want to operate upon. This is how neat image actually works.
NB: I looked at a screen shot. 9 bins, not 10. |
Hmmm... don't know what happened to my reply there...
anyway, you need 9 or more tonal areas in a profile... where does it say you need these tonal areas when sampling a flat area for an individual photo?
Message edited by author 2004-06-15 11:38:20. |
|
|
|
06/15/2004 11:39:09 AM · #42 |
Originally posted by BobsterLobster:
I can see the point of using 10 tonal ranges in a profile, but where does it say that you need 9/10 or more tonal ranges when profiling a flat area in an individual photo?
|
Because you only address the noise that you calibrate the profile for. So no, you don't need them, but you are not doing the best job reducing the noise if you don't profile correctly.
The tool would work with no profiles at all, or a profile based on your randomly moving the sliders around too - it just isn't the best approach for a particular camera.
The point that I keep seemingly am not making well, is that for a given real image, you will not be able to make the best profile, using auto or manual profiling, as you will only sample a sub-section of the tonal ranges you need to consider.
A calibration target gives you the best coverage of all ranges, for a given camera, at a given ISO. Image specific fine tuning can be used to tweak this inital profile to take in to account temperature effects, but it is unrealistic for most images to be able to build a complete profile from scratch, which still appears to be what you are advocating (and is exactly the opposite of what all the NITeam posts on the forums actually say)
Selecting a large patch of sky for example will give you a profile that removes noise from the large patch of sky, at that particular tonal range - and ignores all the other noise in the image (shadow noise etc) If you have large regions in those areas too, then you can address that with an image based profile generation (auto or manual)
I just don't take shots that have large, flat regions at all of the tonal ranges needed to generate a good profile. Most people don't either from what I've seen. Device specific profiles, combined with some fine tuning is the best way to both get a comprehensive and specific profile for any given image from a camera.
Message edited by author 2004-06-15 11:45:12.
|
|
|
|
06/15/2004 11:45:54 AM · #43 |
Originally posted by Gordon:
Originally posted by BobsterLobster:
I can see the point of using 10 tonal ranges in a profile, but where does it say that you need 9/10 or more tonal ranges when profiling a flat area in an individual photo?
|
Because you only address the noise that you calibrate the profile for. So no, you don't need them, but you are not doing the best job reducing the noise if you don't profile correctly.
The tool would work with no profiles at all, or a profile based on your randomly moving the sliders around too - it just isn't the best approach for a particular camera.
The point that I keep seemingly am not making well, is that for a given real image, you will not be able to make the best profile, using auto or manual profiling, as you will only sample a sub-section of the tonal ranges you need to consider.
A calibration target gives you the best coverage of all ranges, for a given camera, at a given ISO. Image specific fine tuning can be used to tweak this inital profile to take in to account temperature effects, but it is unrealistic for most images to be able to build a complete profile from scratch, which still appears to be what you are advocating (and is exactly the opposite of what all the NITeam posts on the forums actually say)
Selecting a large patch of sky for example will give you a profile that removes noise from the large patch of sky, at that particular tonal range - and ignores all the other noise in the image (shadow noise etc) If you have large regions in those areas too, then you can address that with an image based profile generation (auto or manual)
I just don't take shots that have large, flat regions at all of the tonal ranges needed to generate a good profile. Most people don't either. |
Yes, but you still haven't said where in the NeatImage manual/website it says that when using the most desirable method of sampling a 100*100 flat area, that this should ideally have 9 or more tonal areas. |
|
|
|
06/15/2004 11:49:59 AM · #44 |
Sorry, you amended your reply that I replied to!
Here are some more responses...
I don't advocate building a profile from a photo. This should be done as recommended in the manual.
And I see where you're coming from about only sampling one type of noise in an individual photo that might contain other kinds of noise, but it does actually work. This is what I've discovered through experience and trial and error.
I've only looked at the manual today, but it does back up everything I'm saying.
I agree that if you build a profile, you want as many tonal ranges as possible. But the manual does say that using this profile is usually not the optimum way of using NeatImage.
Right, I'm off to live my life again now! :-) |
|
|
|
06/15/2004 11:59:27 AM · #45 |
Originally posted by BobsterLobster:
Sorry, you amended your reply that I replied to!
Here are some more responses...
I don't advocate building a profile from a photo. This should be done as recommended in the manual.
And I see where you're coming from about only sampling one type of noise in an individual photo that might contain other kinds of noise, but it does actually work. This is what I've discovered through experience and trial and error.
I've only looked at the manual today, but it does back up everything I'm saying.
I agree that if you build a profile, you want as many tonal ranges as possible. But the manual does say that using this profile is usually not the optimum way of using NeatImage.
Right, I'm off to live my life again now! :-) |
Again with the missing the point :)
You are always building a profile.
That is the basic point in all this. Neat image always uses a noise profile to perform noise reduction. If you just load an image, and select a large area - you are building a profile.
It is the quickest, easiest way to use the tool and it works quite well. It just isn't the best way to use it. Note also that the on-line version of the manual doesn't even talk about fine tuneing a rough noise profile you generate from an image, although the tutorials spend a lot of time talking about this - it comes down to an ease of use vs quality of results thing.
Message edited by author 2004-06-15 12:06:33.
|
|
|
|
06/15/2004 12:10:48 PM · #46 |
Originally posted by Gordon:
If you just load an image, and select a large area - you are building a profile.
It is the quickest, easiest way to use the tool and it works quite well. It just isn't the best way to use it. |
Everywhere on the NeatImage website says otherwise. It has nothing to do with ease of use, it has to do with picture quality. |
|
|
|
06/15/2004 12:23:29 PM · #47 |
Originally posted by BobsterLobster:
Originally posted by Gordon:
If you just load an image, and select a large area - you are building a profile.
It is the quickest, easiest way to use the tool and it works quite well. It just isn't the best way to use it. |
Everywhere on the NeatImage website says otherwise. It has nothing to do with ease of use, it has to do with picture quality. |
As I said - it is hard to find information on the web site about fine tuning profiles - although all of their staff repeatedly suggest using a profile then doing additional fine tuning, in their forum responses. Most of this is documented in the non-HTML version of the manual.
If you are happy with the results you get, I wouldn't worry about it though. Like you said, some people are just less fussy than others about this ;)
As a final point, here are some other posts on the NI Forum that concur with the approach I am advocating.
Link to Forum
Message edited by author 2004-06-15 12:34:17.
|
|
|
|
06/15/2004 01:40:45 PM · #48 |
| Gordon, I think you ought to run for office. |
|
|
|
06/15/2004 01:45:42 PM · #49 |
Originally posted by coolhar:
Gordon, I think you ought to run for office. |
Anything to get me away from here ? :)
|
|
|
|
06/15/2004 02:26:15 PM · #50 |
Regarding when to use and when not to use neat image...
Neat image can do an awful lot to help smooth and flatter skin tones. I rarely ever use it, though I do play with it to see if a shot could benefit from it.
Here is an example where it smoothed the skin and think it added a lot to the shot: (Great for flattering women)
Noise reduction is real touchy without losing the detail we work so hard to preserve in our shots. Then again, when I invest in a "real" dSLR instead of my Olympus E-10, life will be much easier. For now, the E-10 will have to do.
|
|
