Jump to content

Towards an explanation of the Italian Flag Phenomenon


Lindolfi

Recommended Posts

Pico,

That’s interesting because I don’t know what this is all about anyway. Never seen it,

and, if I had, I’d try and see what I could DO about it. (given the photo was a ‘’ keeper’’).

Back in the film days one had to do their own developing and printing to get results

worth talking about. Now, as it seems to me, the camera is supposed to do everything

for you. Cameras never did and never will.

Link to post
Share on other sites

Lindolfi, I discovered this thread this morning. I have been thinking along the same lines as you, but as in your case, all easy explanations are ruled out by the fact of asymmetry. Light dispersion in the microlenses e.g. should produce the same colour shift right and left.

 

Now if we follow the light all the way from the front lens to the bottom of the sensor, we find that everything that the light encounters and passes on its path is rotationally symmetrical: Lens elements, cover glass (the thickness I have heard about is 0.8mm, btw), the pixel wells – everything EXCEPT the Bayer filter layer. So by exclusion, this is the most probable culprit.

 

– The curious part is the dog's barking.

– But Holmes, the dog didn't bark!

– Exactly Watson. That's the curious part.

 

So, the Bayer filter is the only dog that isn't barking. But the actual micro-causation of it (pardon the neologism) is something of a mystery to me.

 

The Gnomes of Solms have been keeping studiously mum about the matter, that's true. The suggestion that they have a good grip on the phenomenon but won't tell us because [insert favourite explanation here], smells uncomfortably like a conspiracy theory. And that is shrink stuff. But as a historian by training, I am uncomfortably aware that history is full of conspiracies, and even more cases where a conspiracy is not even needed, because everybody concerned knows where his interests lie. And no matter what the Gnomes know, or don't know, it's self-evident that they can't say anything until they have not only an explanation, but also a fix ready for delivery within a week or so. If they, or anybody on this thread, would come up with a credible explanation but no fix, M9 sales would be killed off completely within a few days. And with them, Leica Camera.

 

The old Gnome from the Age of Film

  • Like 1
Link to post
Share on other sites

I could think of a few other dogs.

 

So far, I kept thinking of the sensor plus filter plus microlens as a monolithic unit. That's most certainly not the case.

 

I think that what we perceive as "the sensor" ought to be seen as at least four different layers, from back to front:

  • the actual sensing device with the light sensitive areas, a large IC
  • a layer of coloured bits of matter which pass light of selected colours only - the Bayer mosaic
  • a layer of lenses of a diameter on the same order of magnitude as the individual light sensitive spots and the Bayer mosaic elements
  • a sheet of - presumably coated - glass acting both as mechanical protection and as IR blocking filter.

 

Now I know that within the semiconductor (the actual sensing device) all components are grown on and etched from the same die. Hence, the geometrical positions of all elements ought to be defined to very close tolerances.

 

What I do not know how the Bayer mosaic and the micro lenses arrive at their respective positions. However, I rather think that those are things that are first produced separately and then placed on top of the sensor chip. Observing cameras with different behaviour with respect to the flag syndrome would seem consistent with small deviations of the placement and/or size of one or both of those layers on top of the sensors.

 

I see no way to prove or disprove this so far.

Link to post
Share on other sites

Thanks so much for all your thoughts. Had a great time reading them. To be clear about my own motives: I am not worried at all about the Italian Flag Phenomenon ( or Red Edge) in terms of using the M9 or in terms of the value for me as a camera. As pointed out, most of the time you don't see it with coded or hand dialed lenses and if you do see it, there is cornerfix or other postprocess solutions. But since speculations about the new firmware raised the issue again, as a scientist with a keen interest in optics, I simply can not leave something like that unexplained.

 

Even if some Leica expert comes up with an explanation, it will only be useful if it is detailed, contains cause and effect and is able to account for the variations we see. With variations I mean measurable shifts in colour as a result of changes in lightray angle distribution over the sensor. Another type of variation I would find very interesting to measure is that between M 9 bodies with the same lens mounted.

 

Come to think of it: it may be useful to to mount a pinhole on the M9 at varying distances and measure the colour shift, so that the light ray bundles offered to the sensor is a lot less complicated.

 

Anyway, I hope I have made clear thar I am not trying to raise concern about the M9, just curiosity.

  • Like 2
Link to post
Share on other sites

Advertisement (gone after registration)

How do we know, BTW, that the cyan or red tint is caused by an excess of cyan or red light or by lack of red or cyan light, respectively? Can we be sure that we're not looking at "white" light directly arriving on read, blue or green pits?

 

You wouldn't have the equipment to repeat the first shot with monochrome light, by any chance?

Link to post
Share on other sites

 

I'm starting this thread with the only purpose of finding such an explanation. Other discussions in this thread should be avoided in order to keep momentum in the direction of an explanation.

 

 

 

 

This is a constructive and useful thread. Derailing posts will be deleted from now on.

Edited by jaapv
Link to post
Share on other sites

How do we know, BTW, that the cyan or red tint is caused by an excess of cyan or red light or by lack of red or cyan light, respectively? Can we be sure that we're not looking at "white" light directly arriving on read, blue or green pits?

Assuming that the Bayer filter array plays any role here and that the incident angle is the critical factor (and it has been argued before that we can exclude anything else since the effect would have to exhibit rotational symmetry then), there are two scenarios.

 

1. After a ray of light has passed the red, green, or blue filter, it hits the photosite not of the pixel immediately underneath but of an adjacent pixel – one that is nominally sensitive to a different colour that the filter the ray has passed through. Since each photosite is panchromatic if it weren’t for the filter, it will always register light as being of its nominal colour, i.e. if it is a green pixel its photo electrons will count towards green, even when it is hit by red light after having passed a neighbouring pixel’s red filter.

 

2. Due to refraction in the cover glass (its thickness is 0.5 mm, btw, or 0.7 mm in the case of the M8) a ray of light hits the wrong pixel, both filter and photosite. The colour detected would be correct, just the position at which it is detected would be slightly off.

 

It seems that we can can exclude scenario 2 since its only effect would be a lack of sharpness; there wouldn’t be any discolouration. So it has to be scenario 1. BUT: How exactly would such an effect account for the observed discolouration? That’s the difficult part.

 

Unless of course there is another, entirely different explanation.

Edited by mjh
Link to post
Share on other sites

But would these scenarios not create fringing instead of discoloration?

The second one would not but if the first one did provide any explanation at all it would create colour artifacts on a microscopic (i.e. fringing) as well as a macroscopic scale (colour shift). Fringing would only occur if there was fine detail (of a sufficiently high contrast) in the corners of the image while all the test shots for red corners show some uniformly lit plane without any high-frequency detail. Now if there were images with high frequency detail near the corners and (a) red or cyan corners but (B) no fringing, we could conclude that scenario 1 had to be excluded as well.

 

Which would imply that we’ve run out of explanations.

 

Now it is not like this would surprise me. I don’t see how scenario 1 could even explain the Italian flag phenomenon; it only looks promising until you start to look more closely.

Edited by mjh
  • Like 2
Link to post
Share on other sites

The problems I still have with the "Bayer pattern" theory are these:

 

Bayer 4-pixel units don't exist by themselves, they form a continuous surface. If the light that came through a red Bayer filter falls on a green pixel, due to the angle the light is traveling, then presumably what is hitting the red pixel itself is light that came through another green pixel. (see top half of diagram) And so on. That would appear to cancel out.

 

You'd get a gross color error (red things turning green) but not a subtle gradation.

 

Pixels are tiny, so the difference in angle of incidence from one pixel to the next is extremely small, even with troublesome rangefinder optics.

 

The half-diagonal of the M9 sensor, from center to corner, is 2606 pixels. So with a lens that projects light at a 45° angle of incidence in the corners (pretty extreme), the difference in the angle between two neighboring pixels is at most 45°/2606, or ~0.017 degrees. Virtually parallel for any 2-3 pixel group. Put another way, for the incident light angle to change by 1°, one has to move 58 pixels.

 

It also assumes that there is a gap between the color filters and the underlying silicon such that light can "leak" underneath the Bayer filters to a neighboring pixel. I'm not sure that is really the case.

 

Bottom half of diagram and...http://www.imaging-resource.com/NPICS1/SONY_BACKILLUM_CMOS_2_S.JPG

 

But as Lars says, the effect is not symmetrical, and the only obvious thing not perfectly symmetrical in the imaging path is the Bayer overlay. Plus, as mentioned, Stefan Daniels raised the issue of Bayer asymmetry as at least a factor when he spoke about the problem last fall and said Leica was aiming for improved firmware corrections this spring (now verging on summer - but who's counting? ;) )

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

Link to post
Share on other sites

Bayer 4-pixel units don't exist by themselves, they form a continuous surface. If the light that came through a red Bayer filter falls on a green pixel, due to the angle the light is traveling, then presumably what is hitting the red pixel itself is light that came through another green pixel. (see top half of diagram) And so on. That would appear to cancel out.

Yes, it would tend to cancel out, but not completely. If you take into account the filter characteristics, the quantum efficiency of red-, green-, and blue-sensitive pixels isn’t uniform. Usually green pixels are the most sensitive; blue pixels are usually the least sensitive. This is complicated by the fact that part of the wavelength dependency is due to the silicon’s intrinsic wavelength dependency and the IR absorption filter (that also absorbs red). In the case of the M9 we can probably assume that the combined effect of the characteristics of the Bayer filter array, the IR absorbing cover glass, and the silicon is such that red and blue pixels have roughly the the same sensitivity but are less sensitive than the green pixels.

 

This difference in the quantum efficiency is compensated by adjusting the signals coming from red, green, and blue pixels. The signals from red or blue pixels get amplified more than those of the green pixels, either by an actual (analog) amplification or (more likely) by a correction applied to the digitized raw data.

 

If we assume, just for the sake of the argument, that the differences in quantum efficiency were entirely due to the filter characteristics, then the effect of light leaking to adjacent pixels after having passed the filter array wouldn’t cancel out. Green filters would pass more light than red filters, yet the stronger green light would hit a nominally red photosite and thus get an additional boost; on the other hand the weaker red light would hit a nominally green photosite and receive no boost. The net effect would be the appearance of more red than green. Since roughly the same would happen between blue and green, we should expect a colour shift towards magenta (red + blue).

 

Obviously the real effect would be more complex since the filter array is decidedly not the only factor in the RGB pixels’ wavelength dependency.

 

(Note that this explanation doesn’t get us anywhere since the proposed effect would be the same on all sides of the sensor which is contrary to observations.)

Edited by mjh
Link to post
Share on other sites

Unless of course there is another, entirely different explanation.

 

I think there is.

 

So far we have only considered the optical geometry of light, and wavelength to the extent that we have RGB filters in a Bayer pattern. Obviously we are struggling to explain what we observe.

 

In a way we have also considered the quantum property of light in connection with the detection in the sensor substrate.

 

However, there are two other properties of light that we haven't explored, namely phase and orientation.

 

Consider the asymmetrical shape in the images from Lindolfi, and in particular that they exhibit coloration. The most common cause of these kinds of symptoms is when we are dealing with anisotropic media. What if the micro lenses are birefringent? Or, what if micro lenses, cover glass and lens elements are to various degree strain birefringent? Certainly then you would expect to see variation from different combinations of lens and camera.

 

Birefringence is well known in microscopy, and that's how I came up with the idea because I studied the basics of polarized-light microscopy a few years ago. Actually, you can buy special "strain free" microscope objectives at a cost ;)

 

So, to disprove my "theory" we should set up some clever experiments involving polarizing filters. Notice that I'm a bit fuzzy on the detail; that's because I just realised that I need to refresh my knowledge on the subject :o

 

Anyway, since Lindolfi et al are set up with Matlab and eager to apply their engineering skills I'm sure we can work something out.

 

Regards

Per

Link to post
Share on other sites

Perb's ideas in #35 about phase and orientation of light surprise me as they might give a valuable explanation for my completely unscientific observations about the phenomenon. As far as I can tell, it depends on the "quality" of light, whether the "red edges" are to be seen or not: in situations of clear sunshine - when a polarizing filter might have it's maximum effect - I don't recognize any traces of red/magenta in the edges. In situations of a cloudy sky or even more with artificial light, I can easily see and reproduce this effect. In stead of clouds, an opaque surface - like in landolfi's examples - will work as well.

 

I described this earlier here: http://www.l-camera-forum.com/leica-forum/leica-m9-forum/153831-m9-red-edge-redux-part-ii-2.html#post1540181

Edited by UliWer
Link to post
Share on other sites

Italian Flag Syndrome ???

Oh, when are you guy’s gonna quit picking on Luigi? :)

 

It was meant positively because we speak of Italian Flag Phenomenon: Luigi is a phenomenon indeed

(oops , sorry for being of topic, read it later)

Edited by otto.f
Link to post
Share on other sites

Some observations I would like to see filled in:

 

i] If you take the area covered by the M8 sensor out of an M9 image, using the same lens, and you use no coding, do you get the same amount of colour change? I have no M8, so can not do the test. I'm asking this to see what the effect is of the thickness of the cover glass.

 

 

My 18/3.4 on M8 gives comparable colorshifts, if not corrected.

Edited by otto.f
Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...