Beni wrote:
I'm a wedding photographer using 5Dc's, I get moire every single wedding from the bridal veil. Not terrible but certainly there. As a wedding photographer I wouldn't touch the 800E with a bargepole, the advantages are outweighed for this profession by the downsides. We aren't printing as large as fashion shoots are neither are we lighting as hard, etc.
out of curiosity, are you just talking about the moire or would the large file size be a bigger issue. the D800 should be an improvement over your 5Dc in lowlight, frame rate, and focus speed.
Since the reason was pretty obvious (blue curtains) and the fact that I know the green channel in the D7000 is pretty straight into Adobe RGB I just used the green for luminance, and used a color average as a color blend layer.
The D7000 is average. The D3x is slightly weaker, as far as I've been able to tell. But to make sure, I'd have to run a mask test on the D7000 too, and that's a right bother.
theSuede wrote:
Since the reason was pretty obvious (blue curtains) and the fact that I know the green channel in the D7000 is pretty straight into Adobe RGB I just used the green for luminance, and used a color average as a color blend layer.
The D7000 is average. The D3x is slightly weaker, as far as I've been able to tell. But to make sure, I'd have to run a mask test on the D7000 too, and that's a right bother.
Thanks. Of course, I don't know what color the curtains really are but I get a sense that they were more blue (top right of the pre-corrected example) and the color average technique has maybe averaged the red/magenta and blue to come up with a more purple hue? If so, easy fix for that anyway.
Beni wrote:
I'm a wedding photographer using 5Dc's, I get moire every single wedding from the bridal veil. Not terrible but certainly there. As a wedding photographer I wouldn't touch the 800E with a bargepole, the advantages are outweighed for this profession by the downsides. We aren't printing as large as fashion shoots are neither are we lighting as hard, etc.
Beni, it depends what final resolution you need. For example, using RPP's halfsize, you would get a 9MP image with perfect colour information for every pixel. I imagine that moire would not be an issue then.
carstenw wrote:
Beni, it depends what final resolution you need. For example, using RPP's halfsize, you would get a 9MP image with perfect colour information for every pixel. I imagine that moire would not be an issue then.
If you get bad moire in an image, such as a wedding veil for instance, it will be readily visible even at smaller sizes from everything I have seen unless it's minimized in pp.
Wedding photographers (and fashion photographers for that matter), would do better with an AA filter because the PP required to deal with severe moire when it shows up is a major pain.
Besides the color morie issue, there are other pattern issues such as this example:
One thing about moire vs. additional sharpening ... the moire is ALWAYS a selective issue rather than a global one. Throwing a little more USM at an image that is shot on an AA filter is a GLOBAL issue. Correcting moire' is not.
Global issues are WAY EAISER to deal with than selective issues ... you can batch/action deal with sharpening. Each image with moire must be dealt with individually and selectively.
carstenw wrote:
Beni, it depends what final resolution you need. For example, using RPP's halfsize, you would get a 9MP image with perfect colour information for every pixel. I imagine that moire would not be an issue then.
Half-size actually makes it worse, not the other way around...
Converting using half-size discards 50% of the green resolution, that you desperately need to be able to rebuild the damaged area.
Tariq Gibran wrote:
If you get bad moire in an image, such as a wedding veil for instance, it will be readily visible even at smaller sizes from everything I have seen unless it's minimized in pp.
I am not sure. How much of the moire is a result of debayering with imperfect information? Colour moire would always be, I would think. RPP's half-size mode doesn't debayer, it grabs 4 pixels and has a complete colour. You might still get some normal moire, but I don't think AA filters would avoid this kind.
theSuede wrote:
Half-size actually makes it worse, not the other way around...
Converting using half-size discards 50% of the green resolution, that you desperately need to be able to rebuild the damaged area.
Could you explain that more? I am not sure I get why. RPP's half mode doesn't debayer.
carstenw wrote:
I am not sure. How much of the moire is a result of debayering with imperfect information? Colour moire would always be, I would think. RPP's half-size mode doesn't debayer, it grabs 4 pixels and has a complete colour. You might still get some normal moire, but I don't think AA filters would avoid this kind.
I'm not following...you must "debayer" with every raw converter at some point in order to get the image, right? and, as you say, the moire is caused by imperfect information during initial capture.
No, RPP doesn't debayer in its half mode. It just grabs four pixels and adds them directly, i.e. it grabs the RGBG information and adds it. 36MP would then yield a 9MP image, but with perfect colour information.
Sure.
The "bin" mode just takes a 2x2 region of [R][G][G][B] pixels and makes a single [RGB] pixel out of it. This means that every last bit of aliasing (that is mostly present in the R and B channels) is still there in full strength the 50% size result. AND as icing on the cake you have also reduced two greens into one by averaging, giving no heed to their spatial relationship. You reduce spatial resolution by 50%, and keep all the moire in.
The moire is not an interpolation ("de-Bayer") artefact, a "fault" in the interpolation. It is present in the raw file as a result of the Red and Blue pixels having large empty spaces around them. Every red and blue pixel has eight "voids" surrounding it, areas where the camera has no idea what the red/blue target content is. Given no understandable clue, the guesstimate can only be "wrong" or "more wrong". Well, it can also be "totally off the chart wrong", but that's rare.
RustyBug wrote:
One thing about moire vs. additional sharpening ... the moire is ALWAYS a selective issue rather than a global one. Throwing a little more USM at an image that is shot on an AA filter is a GLOBAL issue. Correcting moire' is not.
Global issues are WAY EAISER to deal with than selective issues ... you can batch/action deal with sharpening. Each image with moire must be dealt with individually and selectively.
This is really only true if you sitting staring at it in PS. It's not really true if you're tasked with creating a filter to remove it. The latter is a global change in that it looks at the entire image, although it could be selectively applied using a mask of course. But it needs to do a complex DFT of the image since moiré shows up as spikes in the (2D) image frequency; the filter locates (possibly by correlation or heuristically) the spikes and attenuates them, leaving the phase (and hence edge response) unaltered. This is something you can't easily accomplish with the end-user tools provided in PS. (With matlab or mathematica it's pretty easy. And that's how you'd prototype it of course, then go off and create the production filter.)
carstenw wrote:
No, RPP doesn't debayer in its half mode. It just grabs four pixels and adds them directly, i.e. it grabs the RGBG information and adds it. 36MP would then yield a 9MP image, but with perfect colour information.
I don't know about that but even if it was the case/ possible, the four pixels RPP are grabbing have already been contaminated by the error during initial capture which causes the moire - that initial capture error is not avoided and it's effect is often more widespread than just four pixels.
I still suspect that what RPP actually is doing is simply interpolated the raw converted image down in size as I don't see how the image could even exist as an image without going through the de-mosaic process. If this is somehow possible - and RPP is capable of it - I would love to hear about it from Joakim.
RPP is capable, as would any other raw program be, if they should choose to do so. The information exists as one colour channel per pixel already, you just need to not fill in the gaps Most raw formats that I know of are simple variants of TIFF, so they are not actually that difficult to understand.
Jan Brittenson wrote:
This is really only true if you sitting staring at it in PS. It's not really true if you're tasked with creating a filter to remove it. The latter is a global change in that it looks at the entire image, although it could be selectively applied using a mask of course. But it needs to do a complex DFT of the image since moiré shows up as spikes in the (2D) image frequency; the filter locates (possibly by correlation or heuristically) the spikes and attenuates them, leaving the phase (and hence edge response) unaltered. This is something you can't easily accomplish with the end-user tools provided in PS. (With matlab or mathematica it's pretty easy. And that's how you'd prototype it of course, then go off and create the production filter.) ...Show more →
And I think you just made a compelling case that for the "average" user (who isn't using matlab & mathematica) who has PS or LR as their tools to use for dealing with moire' ... they really ought to think about moire' and non-AA with a bit more credence to what it means to their workflow.
I realize that we are "alt" users and like to "tinker" ... but would one need to develop a different filter prototype for each len's unique resolution and aperture combination ... i.e. one for a 14-24G and a different one for a ZF 25/2 and yet another one for a 100/2 Planar, etc. ... in order to have a 'preset' solution ?? Or would it be a "one size fits all" ?? And if it's a "one size fits all", why hasn't it already been produced commercially as a plug-in ??
Jan Brittenson wrote:
This is really only true if you sitting staring at it in PS. It's not really true if you're tasked with creating a filter to remove it. The latter is a global change in that it looks at the entire image, although it could be selectively applied using a mask of course. But it needs to do a complex DFT of the image since moiré shows up as spikes in the (2D) image frequency; the filter locates (possibly by correlation or heuristically) the spikes and attenuates them, leaving the phase (and hence edge response) unaltered. This is something you can't easily accomplish with the end-user tools provided in PS. (With matlab or mathematica it's pretty easy. And that's how you'd prototype it of course, then go off and create the production filter.) ...Show more →
Been there, done that, written the commercial video plugins. No, it doesn't work in that easygoing way...
*The correlations is always unknown, since the algorithm doesn't know if the moire is caused by blue aliasing or red aliasing, or a combination.
*The DFT (FFT) of the image only shows spikes if the moire is present on a perfectly flat surface covered in a perfectly repetitive (and uniform) pattern. Take a wavy veil over a bride - Where's the perfect repetition that would give a spike?
Try an FFT on the image in that was linked earlier, on photo.net.
carstenw wrote:
No, RPP doesn't debayer in its half mode. It just grabs four pixels and adds them directly, i.e. it grabs the RGBG information and adds it. 36MP would then yield a 9MP image, but with perfect colour information.
Err. No. This won't give you perfect chroma. To get the chroma you need to low-pass filter it.
Binning really does suck and gives you lots of artifacts, which ironically may include moiré.
theSuede wrote:
Take a wavy veil over a bride - Where's the perfect repetition that would give a spike?
The image clearly shows a 2x progression in frequencies - if you take a ruler to the lines you'll see they're separated by fixed distances. This shows as a series of frequency spikes at 2x distance.
However, this image isn't fixable as shown because someone tried to remove the effects by desaturating specific colors. You need the original data.
You can look for it in the luminance or xy in Lxy. Doesn't terribly matter.
Most raw formats that I know of are simple variants of TIFF, so they are not actually that difficult to understand.