7D Maze/low ISO artifacts UPDATE

John Sheehy · Oct 28, 2009

ejmartin said:
John Sheehy said:

You wouldn't correct anything unless the local means were off. What is the chance that the subject aligns with the lines?

Click to expand...

ejmartin said:
You mean like this (D70 shot)?

The D70 is notorious for its almost non-existent AA filter, on only 6 MP. The chances of getting that on a 7D are less, and will be even less with future sensors with higher densities.

ejmartin said:
The columns between the windows are Nyquist level texture. If they are to be resolved, the greens must be matched. Any averaging of the greens gives mush. While this is an extreme example, it illustrates the fact that averaging the greens is going to drop the resolution of the camera, no matter how the texture is oriented.

I have never advocated averaging of greens for the 7D. In fact, it is the worst solution (although, with oversampling, it would not be an issue). What I suggested was that the low frequency components be altered, and except for horizontal or vertical high-contrast repeating patterns at the nyquist, the simple version of the conversion solution would not lose detail at the nyquist. The problem we're worried about, mazing in flat, saturated non-green, colored areas, only need to be corrected in such areas; not applied to gray parallel lines. It should probably be included as an optional feature in converters.

ejmartin said:
ejmartin said:

Another variation on what I wrote would be to get the smoothed versions of each channel, average them together, and then add back the difference between the literal channels and their individually smoothed versions, so effectively, all you are doing is pulling the low-pass versions of the two green channels together, without affecting the high-pass versions of the individual channels.

Click to expand...

Any such averaging throws away data that allows the highest resolution that the sensor array is capable of, if the greens were matched.

Only with lines at the nyquist - even if applied globally, the methods I mentioned would only cancel detail near the nyquist when the means of greens in lines is at the nyquist.

ejmartin said:
ejmartin said:

If the response is consistent across all specimens, the local color could be used to force a local adjustment. There are a number of ways to deal with this.

Obviously, correction is not going to be easy in area of high chromatic frequency, but then again, you probably would not notice the mazing there, anyway.

Click to expand...

ejmartin said:
Look, there are ways of dealing with the mismatch. The point is that to do so means turning an 18MP camera into a substantially lower MP camera. The green channels carry the highest spatial frequency luminance information, and to have to average them amounts to low pass filtering that information. It is lost to the demosaic process.

Yes, but only in special situations, when applied globally, without condition.

--
John

grammieb14 · Oct 28, 2009

This is what the program looks like without a photo downloaded to it.{

}

GaborSch · Oct 28, 2009

1. DPReview measures the DR on JPEG images. This does not reflect the cameras' capabilities but the rendering engines' pecularities.

2. The published numbers are often in the realm of jokes, even with JPEGs.

Example: the 5D2 with ISO 1600: 8.4 EV, with ISO 3200: 8.6 EV. In fact, the DR with 3200 is 1 EV lower than with 1600.

Example: the 40D with ISO 1600: 8.9 EV, with ISO 3200: 7.4 EV, i.e. the loss is 1.5 EV with only 1 EV exposure difference .

This is plain garbage . I am basing my statements on my measurements of raw (i.e. non-converted) image data.

--
Gabor

http://www.panopeeper.com/panorama/pano.htm

Julio · Oct 28, 2009

I'm trying but their servers are overwhelmed and I won't be able to do anything with it for several hours anyway.

It has full 7D support so I'll be very curious to see how it deals with the g1/g2 imbalance, particualrly with your 7D.
--
http://fotoman99.smugmug.com/

bronxbombers · Oct 28, 2009

grammieb14 said:
This is what the program looks like without a photo downloaded to it.{

}

thanks, it looks like perhaps those two green sliders are related to the two different green channels, but i'm not really sure what is measures when an image loads, i'd have to look at the manual, maybe they have it online

bronxbombers · Oct 28, 2009

no, i will try the demo but the servers are still too busy

Mark H · Oct 28, 2009

John Sheehy said:
OK; I did a quick test with a narrow-band red LED, separating the channels in IRIS after clipping away the masked borders:

R: 6253.0
G1: 1164.7
G2: 1197.3
B: 190.4

That's a difference of about 2.8%

Then, a blue LED ($2 more):

R: 47.9
G1: 1175.5
G2: 1156.5
B: 3686.0

about 1.6%, in the opposite direction.

Took a second blue LED shot to see if the shot varies any gain difference, but the difference was almost exactly the same.

So, my guess is that there is some small, and maybe unintentional difference in spectral response between the green channels.

This is almost exactly the same test that I described here...
http://forums.dpreview.com/forums/read.asp?forum=1019&message=33359097
...except I use a PC monitor as a colour source rather than LEDs.

What is curious though, is that in your results the higher green channel is in the opposite column to my Dimage 7 camera's behaviour.

Are you sure your G1 G2 designations are correct i.e. G1=red column and G2=blue column?

John Sheehy · Oct 29, 2009

Mark H said:
This is almost exactly the same test that I described here...
http://forums.dpreview.com/forums/read.asp?forum=1019&message=33359097
...except I use a PC monitor as a colour source rather than LEDs.

I figured a narrow-band LED would contribute none of the green which might hide the difference.

Mark H said:
What is curious though, is that in your results the higher green channel is in the opposite column to my Dimage 7 camera's behaviour.

Are you sure your G1 G2 designations are correct i.e. G1=red column and G2=blue column?

In Iris, "split_cfa r g1 g2 b" assigns the channels based on left to right on the bottom left corner, and then the next line up, left to right, like so:

|
| 3 4
| 1 2
----------

--
John

erikstefan · Oct 29, 2009

According to this Blog it's the RED channel has banding

http://diglloyd.com/diglloyd/free/PushingTheBlacks/index.html

GaborSch · Oct 29, 2009

Not channels but columns cause the banding, and in each column two channels are affected. The banding is apparent in either all four channels, or in the red and one green, or in the blue and the other green.

It's important to understand, that the issue is not the green balance . It is misguided to search for clues and solutions by playing with the greens.

--
Gabor

http://www.panopeeper.com/panorama/pano.htm

ejmartin · Oct 29, 2009

GaborSch said:
Not channels but columns cause the banding, and in each column two channels are affected. The banding is apparent in either all four channels, or in the red and one green, or in the blue and the other green.

It's important to understand, that the issue is not the green balance . It is misguided to search for clues and solutions by playing with the greens.

Sorry, there is not a single issue under discussion . You are obsessed about column offsets near black; other people are discussing gain imbalance which affects all tonalities, especially upper tones, for which playing with the greens is of the essence.

--
emil
--

http://theory.uchicago.edu/~ejm/pix/20d/

GaborSch · Oct 29, 2009

ejmartin said:
Sorry, there is not a single issue under discussion . You are obsessed about column offsets near black; other people are discussing gain imbalance which affects all tonalities, especially upper tones, for which playing with the greens is of the essence.

You are obsessed with the green imbalance; however, it is not proven, that that is not the product of the imbalance between the columns (though I don't see any proof of the opposite either).

IMO misaligned filters are not the reason for the green imbalance.

--
Gabor

http://www.panopeeper.com/panorama/pano.htm

ejmartin · Oct 29, 2009

GaborSch said:
it is not proven, that that is not the product of the imbalance between the columns (though I don't see any proof of the opposite either).

The imbalance is not constant in a column, rather it is correlated with the color.

--
emil
--

http://theory.uchicago.edu/~ejm/pix/20d/

bronxbombers · Oct 29, 2009

GaborSch said:
Not channels but columns cause the banding, and in each column two channels are affected. The banding is apparent in either all four channels, or in the red and one green, or in the blue and the other green.

It's important to understand, that the issue is not the green balance . It is misguided to search for clues and solutions by playing with the greens.

well it depends what you are after, the banding or the mazing/noise/artifacts

--

Big lights will INSPIRE you, Mariano will RETIRE you, let's hear it for New York, New York, New York!!

Mark H · Oct 29, 2009

John Sheehy said:
Mark H said:

What is curious though, is that in your results the higher green channel is in the opposite column to my Dimage 7 camera's behaviour.

Are you sure your G1 G2 designations are correct i.e. G1=red column and G2=blue column?

Click to expand...

In Iris, "split_cfa r g1 g2 b" assigns the channels based on left to right on the bottom left corner, and then the next line up, left to right, like so:

|
| 3 4
| 1 2
----------

Unfortunately that wasn't quite as clear as I hoped - normally I'd expect the bottom left pixel '1' to be green (of course there's no one law/rule that it has to be).

However, I'm thinking it suggests your 'G1' is Iris's channel 2, and that is the green of the 'Green-Blue' column?

Am I right?

If that was the case, then your 7D's behaviour would correlate to my Dimage 7's behaviour - which would support my theory/claim in relation to 'cross-talk/signal measurement smearing'.

If I re-order and rename your G1 G2 designations then it would demonstrate how the two green signals are effectively 'pulled' slightly in the direction of the signal level of the other colour (red or blue) in the column that they may share an amplifier-A/D converter with (and the red and blues 'pulled' slightly toward their associated green's signal level)...

John Sheehy said:
R: 6253.0
G(r): 1197.3
G(b): 1164.7
B: 190.4

That's a difference of about 2.8%

Then, a blue LED ($2 more):

R: 47.9
G(r): 1156.5
G(b): 1175.5
B: 3686.0

about 1.6%, in the opposite direction.

John Sheehy · Oct 30, 2009

Mark H said:
John Sheehy said:

Mark H said:

What is curious though, is that in your results the higher green channel is in the opposite column to my Dimage 7 camera's behaviour.

Are you sure your G1 G2 designations are correct i.e. G1=red column and G2=blue column?

Click to expand...

In Iris, "split_cfa r g1 g2 b" assigns the channels based on left to right on the bottom left corner, and then the next line up, left to right, like so:

|
| 3 4
| 1 2
----------

Click to expand...

Unfortunately that wasn't quite as clear as I hoped - normally I'd expect the bottom left pixel '1' to be green (of course there's no one law/rule that it has to be).

However, I'm thinking it suggests your 'G1' is Iris's channel 2, and that is the green of the 'Green-Blue' column?

Am I right?

The user labels the channels in IRIS. It does not pay any attention to different CFA schemes until you do its simple "demosaicing", and even then, it does it by user settings.

You issue the command, "split_cfa a b c d", and the channel in the 1,1 position at the bottom left of the RAW image is "a", 2,1 is "b"; 1,2 is c; 2,2 is d.

The lower left corner of the RAW rectangle extracted from the file by IRIS is:

GB
RG

Mark H said:
If that was the case, then your 7D's behaviour would correlate to my Dimage 7's behaviour - which would support my theory/claim in relation to 'cross-talk/signal measurement smearing'.

Mark H said:
If I re-order and rename your G1 G2 designations then it would demonstrate how the two green signals are effectively 'pulled' slightly in the direction of the signal level of the other colour (red or blue) in the column that they may share an amplifier-A/D converter with (and the red and blues 'pulled' slightly toward their associated green's signal level)...

Assuming a vertical readout, that would match.

--
John

Search

7D Maze/low ISO artifacts UPDATE

John Sheehy

Forum Pro

grammieb14

Senior Member

GaborSch

Veteran Member

Julio

Senior Member

bronxbombers

Forum Pro

bronxbombers

Forum Pro

Mark H

Senior Member

John Sheehy

Forum Pro

erikstefan

Leading Member

GaborSch

Veteran Member

ejmartin

Veteran Member

GaborSch

Veteran Member

ejmartin

Veteran Member

bronxbombers

Forum Pro

Mark H

Senior Member

John Sheehy

Forum Pro

Keyboard shortcuts