[Adrian La-Garde]

Hi Rob,

I was just wondering if you've ever come across a method of sharpening that applies more strongly to red and blues, and less to green and neutrals, to compensate for the varying sensitivity of red, green, and blue sensor pixels in camera?

I have used method this in a printing program called Qimage; the developer of the program discusses his method here: http://www.steves-di...arpness-eq.html

I'm just asking out of curiosity, as I've never seen it mentioned anywhere else, and wondered if you've ever encountered the issue of variable sharpening?


Adrian

[Ray]

Hi Adrian,

I'm obviously not Rob, I guess (and hope) he's busy.
I was thinking maybe not to reply, but this is an open forum, and the following is my opinion, which may or not be similar to others ...

I'm a user of qi too, but only for printing. Mike's article you linked to above is interesting, but I think maybe some 'wrong?' conclusions are drawn. There are three related issues mentioned there.
1) Fovean sensor
2) Bayer sensor colour pixel distribution
3) selective tone sharpening.

So, in reverse order-
3) can be useful as a quick fix for some images. Qi has, afaik, no other selection method - e.g. lasso tool, masking, whatever, and selecting a colour is about as simple as it gets in programming. For images such as a single flamingo on a blue or green background, then it is easy to select and enhance what is presumably the centre of focus. However, in more general (landscape images, say) where similar coloured items are spread throughout the image, then you will be sharpening (or blurring) items that should be left alone.

2) not all cameras have the same Bayer type layout of colour pixels, (but that doesn't effect the nub of the discussion). Also, it makes sense since the human eye is more sensitive to green. The camera manufacturers have teams of guys who in software (and hardware) will be getting the colour balance pretty good. Interestingly, if you examine the examples of the before and after small images (the black spokes on the colour patches in Mike's article) using a colour picker tool, you will see that in the black on red, the black has a large amount of red component in the before image (in areas well within the black). Also in the 'after image' the blue is completely messed up. I do not think this is totally connected with jpeg artefacts, but it looks as if the images have been skewed to make the point of the discussion, but it would be fairly easy for you or any one else to make their own images and test out the theory. As camera hardware improves, the effects may become less noticeable (in the same way as printing at native resolution becomes less noticeable/necessary).

1) The Fovean sensor concept is quite wonderful, being the nearest to colour film in construction, and many of us are used to looking at film images. However, as the light passes through the various layers of silicon, red still becomes defocussed, but generally the amount of blurring is trivial. It is difficult for me to measure and quantify or scientifically prove, the Fovean sensor produces a better quality image, something I can recognise when I see it. It's a pity that Dave Coffin does not cover the full range of Sigma cameras in his raw file processing.

In summary, answering a question I was not asked, I do not think that tone selected sharpening is worth the effort, there are very few images for which it is the best selection method. However, if editing in qi, that is all you get, afaik, so it will be boosted by the qi fans beyond it's true value in the wider world of specific imaging editing software such as Sagelight, where other, more general selection tools are available.

Best wishes,

Ray

[Adrian La-Garde]

Hi Ray,

Thanks for your comments. I, like you, use Qimage for printing as it does an excellent job. Until I found Sagelight I used to also use the Qi editing tools for basic enhancements, including sharpening before saving jpegs. My question was purely for curiosity sake as I did find the Qi colour sharpening often gave better results than a pure USM; however, the effect is more noticeable, as you said, on pictures containing distinct red and green areas. Anyway, using Sagelight's Lightblender Definition/Sharpen layers and vibrance controls adds a lot of "punch" and sharpness to pictures, so I suspect tone sharpening would not actually add much to the overall result.

I was just curious as to whether Rob had come across this method of sharpening since I've never seen it mentioned in any other product.

Even though I only ever use Sagelight now for all my photo editing, I still use Qimage for printing as it does a great job at automatically resizing/resampling/sharpening at print time, and gives a lot of control over the page layout. If Sagelight had this sort of printing functionality, and an in-built file browser, it would be almost perfect - but that's just my preference and Rob never ceases to surprise me with all the new features he keeps adding anyway!!

Adrian

[rob]

Adrian La-Garde, on 17 June 2012 - 09:49 PM, said:

Hi Rob,

I was just wondering if you've ever come across a method of sharpening that applies more strongly to red and blues, and less to green and neutrals, to compensate for the varying sensitivity of red, green, and blue sensor pixels in camera?

I have used method this in a printing program called Qimage; the developer of the program discusses his method here: http://www.steves-di...arpness-eq.html

I'm just asking out of curiosity, as I've never seen it mentioned anywhere else, and wondered if you've ever encountered the issue of variable sharpening?


Adrian

Hi, Adrian.

Sorry for the delay in the reply. I was hoping someone would put in their view of it -- thanks Ray!

I looked into the issue, and have a couple comments, but I also can't comment too much because the article itself didn't really point to any useful examples, and the one presented didn't really make the case.

Not that the thought behind it is bad at all, but I don't think it's necessarily a good generic solution. Or, not that the case isn't good, either, it's just that the article didn't have a practical example to consider.

Also, it seems like the issue of 'tone sharpening' gets mixed up with the idea that the individual colors channels should be sharpened at different levels.

I understand the idea -- that certain colors tend to be blurry because they're only getting the benefit of part of the typical 2x2 ccd area (i.e. GRGB) -- so, if you have something purely red, you're just getting the 'R' component.

I get that, but I would see that as more of an intensity problem, and it is pretty clear that the blue and red channels tend to look less clean and crisp as the green channel. A lot of this, though, has to do with the idea that Green tends to be about 60% of the light, with red about 30% and blue tending to be a little above 10% of the overall light, which leaves the R and B channels more subject to noise.

I don't really like the idea of sharpening the channels differently for a few reasons (again, not that it's a bad idea, I just see drawbacks):

1. Halo/Fringing issues. If you sharpen different channels with a different radius, you could generate fringing on that color channel, as the edge radius would be wider, and you could see this later on in the process, especially if you add Definition/Focus or other sharpening in a generic and overall sense.
2. Noise. The red channel -- and particularly -- the blue channel tend to have a lot of noise in them. In general, a lot of images have low-level noise in shadow areas. If you sharpen these channels more aggressively, more noise can come out in some circumstances.
3. L-channel sharpening. It is better, overall, imo, to do CIE L channel sharpening. This is an option in the Unsharp Mask and is the default for the Definition, Focus, and every other mode that does sharpening in Sagelight. This prevents color noise because it is specifically avoiding sharpening the color, and concentrating on the light values only.

You can try this test with the unsharp mask: try a small radius with a very large strength. Then switch between the C*I*E L channel and RGB sharpening -- with most pictures, you'll see a lot of color noise appear and disappear (between RGB and C*I*E L channel, respectively).

Realistically, I should probably have this as the default option in the Unsharp Mask (right now it's RGB).

I think a lot of the time when we see blurry reds, for example, it can be because of JPEG and other lossy compression issues -- like the stairstepping we see in reds on TV channels. When you compress to a 4:1:1 or 4:2:2 compression (i.e. 4:1:1 means that for every block of 4 pixels, the colors are sampled across 1 pixel for each A & B color component, meaning certain colors (i.e. Reddish) are an average of the 4 pixels for each pixel, and have the same value across these four pixels, leading stairstepping and blurriness), this can cause these sort of issues. I say that just to make the difference between JPEG which causes these problems vs. RAW, where -- though I understand the issue -- haven't really seen this as much of an issue.

I'm not dismissing the idea -- I'd just like to see some examples of it in a setting where the above concerns can also be tested.

With Tone Sharpening, though, which is something else QImage talks about -- I like that from a certain perspective (see below), because it's really about what I see as a truism in image editing -- when you localize an area and work on it -- whether it's sharpening, add or removing color, toning, etc. -- this is what gives an image real depth and allure, when performing the same functions over an entire image can seem overbearing and sometimes gaudy (for example, adding a lot of color to selective areas would add depth, where otherwise it might look too overwhelming).

I know you didn't really talk about the 'Tone Sharpening' aspect of it, but I thought I'd mention it since it seems to get inadvertently tied to the issue of sharpening the RGB channels separately and at different levels.

Tone Sharpening has nothing to do with the issue of sharpening colors with different levels because of CCD issues -- it's just a good overall thing to look it. In general, I would relate Tone Sharpening to something more generic (which seems to be the basis of most of the Tone Sharpening examples anyway): if you sharpen the foreground and leave the background alone (or blur the background), you can really add depth to your image. But, this is more about localizing the foreground object than the actual tone, since if you had something in-between the foreground and background with the same color, this could make a picture look a little off. So,, for me, the idea is more centered on the issue of local 'object sharpening' as opposed to Tone Sharpening -- but a lot of things have the same tone to them, so it can be a nice thing, too.

Rob

[Adrian La-Garde]

Hi Rob,
Thanks for taking time to reply, that was an interesting explanation of some of the issues behind sharpening. As I said to Ray, my question was just out of curiosity as in practise I find I get great sharpening and "punch" to my pictures mostly using Definition and Vibrance tools.
Thanks,Adrian

[Steve Ellsworth]

The article was written circa 2004
hope things have improved by now

i pulled down their samples and ran them through Sl
needless to say i could improve on any of them

the attached file is from one of their users
didnt take much to whip it into shape

i tone map for sharpness much of the time - but usually it doesn't show.

[luc]

Hi all,

this sharpening-by-color is supposed to compensate for the 'unbalanced" sensor design; thus, I would guess this type of sharpening should be of the kind of "capture sharpening" , meaning a rather mild and very small radius initial sharpening? Thus it might not cause bigger halo artifacts when applied to color channels individually?

Why would it be necessary to apply this sharpness compensation to color channels individually - would it not be possible to select/mask preferentially red and blue regions of the picture (automatically) and apply this capture sharpening step to all channels (or of course better the L-channel) but with varying intensity depending on blue/red color content?

[rob]

luc, on 21 June 2012 - 07:29 PM, said:

Hi all,

this sharpening-by-color is supposed to compensate for the 'unbalanced" sensor design; thus, I would guess this type of sharpening should be of the kind of "capture sharpening" , meaning a rather mild and very small radius initial sharpening? Thus it might not cause bigger halo artifacts when applied to color channels individually?

Hi, Luc.

I think the idea is to use a larger radius or strength on the sharpening. In this case, it could cause halo problems. For example, if you have a generally plain area (i.e. not specifically one of the primary RGB colors) with a red object in front of it, such as a flower with the sky or a wall behind it, the area in the radius for the red channel would turn red around a larger edge causing a halo that you may see under some circumstances, especially with further editing.

Capture Sharpening? Can you tell me more about what you mean?

luc, on 21 June 2012 - 07:29 PM, said:

Why would it be necessary to apply this sharpness compensation to color channels individually

<br class="Apple-interchange-newline">
The idea of the article is because of the intensity differential between values based on the color. As I mentioned, I'd need to see some real-world examples, and there were none presented in the article, especially since there are halo and noise considerations, as well as the idea that it tends to replace L-channel sharpening, which is probably better overall.

luc, on 21 June 2012 - 07:29 PM, said:

would it not be possible to select/mask preferentially red and blue regions of the picture (automatically) and apply this capture sharpening step to all channels (or of course better the L-channel) but with varying intensity depending on blue/red color content?

I guess you could, but I think the same problems would crop up. An algorithmically-aggressive way to do it would be to map these areas by contiguous mapping and then 'shrink' these areas, then sharpen to avoid potential halo issues, and I don't know really what you'd end up with, especially if you have background, foreground, and middle objects getting sharpened in all different varations, which could cause a depth-perception confusion in the image unless the different sharpening was extremely precise.

Rob
.

[luc]

Hi Rob,

capture sharpening is part of the idea that often three different types and stages of sharpening might be applied to most images for optimal results: capture sharpening, creative sharpening, and output sharpening.

Please see the explanation halfway down the page here:
http://www.cambridge...-sharpening.htm

[rob]

luc, on 03 July 2012 - 05:36 AM, said:

Hi Rob,

capture sharpening is part of the idea that often three different types and stages of sharpening might be applied to most images for optimal results: capture sharpening, creative sharpening, and output sharpening.

Please see the explanation halfway down the page here:
http://www.cambridge...-sharpening.htm

Hi, Luc.

Thanks for the link. I looked at the article, but I am not sure I caught what you were pointing out. Can you describe it more?

Rob