Tech wonders: Oly's predictive focus for fast AF

goblin said:

Yohan Pamudji said:

Coating matters:
http://dpanswers.com/content/irphoto_lenses.php

Click to expand...

Good point, but what is described there matters for IR photography (image quality, ghosting, flare), not for IR the described IR assisted focusing.

Click to expand...

In infrared photography one often has to deal with internal reflections and hotspots, often due to strong IR reflection from the surface of digital sensors (compared to film in analog cameras).

If there really was an IR-blocking coating on the lenses, they could not be used for infrared photography at all. Reality is that all lenses can be used, but many of them have compromised image quality due to internal reflections and other issues. In general, recent lenses are worse for infrared photography compared to old lenses with simple coatings, due to the recent multicoatings being optimised for visible light. But the new coatings certainly don't block the infrared.

Rriley said:

Iforgetwhat8was4 said:

Ah, that makes some sense, even if it's not "predictive". Although "as sharp" could be a real challenge if the IR contrast did not match up well with the visible light contrast...

Click to expand...

that's how it works, it depends on IR contrast not equalling visible spectrum. This is the same with colours of light, and that's why we have CA in lenses. Each wavelength of light has a different RI (refractive index), but IR spectrum light is quite far from the colour separation RI, that makes it a better indicator

Once you establish that one lot of contrast is less than the other the system knows which way to travel to obtain focus and has a predictor as to where that point is. Or if you like when visible spectrum focus becomes on-target, the IR focus being off-target is guaranteed

Click to expand...

yes, that is my guess as well (haven't read the patent). IR always focuses 'closer' than visible (remember the IR mark on old lenses), so you know the direction required for attaining best focus. The problem is that the amount of visible-IR focus shift varies depending on focal length and lens design, but I guess they have found a solution for that e.g. by storing the parameters of their lenses.

anyway, looks interesting. I wonder what this all means for IR shooters, do they have 'IR pixels' on the chip that we could use for IR imaging, and how does that work with the IR blocking filter? Will the new lenses have worse performance for IR photography?

trmz said:

Since it's contrast what you're looking for, you need to get it somehow: in IR, if you try to focus onto something "warmer" than the surroundings, it's probably ok. But for sure you'll find some situation where the contrast is poor even at those wavelengths, or even poorer than what it is in visible light.

Click to expand...

temperature only becomes relevant above 400 degrees centigrade or so for these wavelengths, to it is not an issue. But yes, some subjects will have very poor contrast in IR while they have good contrast in visible. People who are mostly shooting test charts might be in for rough times

dbateman said:

How olympus could have implemented this is to leave the IR Cut filter on the sensor. ...

Click to expand...

thanks for the good analysis, I should have read that before replying higher up in the thread. Confirms most of my thoughts, we will have to wait and see what they are really doing regarding the IR cut filter and lens coatings.

dbateman said:

The only fiter that IR does not pass through is the BW 486 filter. It filters out UV and IR the best.

Click to expand...

there are several others of course although with a less sharp cutoff, like the B+W 489 and Schott BG38. Even some ND filters (depending on brand) have significant absorption in the 700-1100nm infrared band.

Rriley said:

Thinking further, if they are manipulating the spectrum of IR light that comes through the lens, you might be able to duplicate that with an IR Cut filter screwed onto the lens for existing Zuikos, much the same way M8 operates. That would be intended to replace the coatings theyre using on the new lenses. I cant help thinking there has to be more to it than this though.

Click to expand...

why do you think Olympus is manipulating the IR spectrum with lens coatings? If IR doesn't come through the lens you cannot use it for this IR AF system. Or do you suggest the system uses e.g. the shorter IR wavelengths (like 700-900nm) for fast focusing, and removes the longer IR wavelengths because of adverse effects on the AF or colors in visible imaging?

I don't think you can compare with the M8, because its problem is simply that its internal IR cut filter is too thin (weak), requiring an external IR cut for many normal lighting situations. The M8 doesn't need IR light at all to function, unless you use it for IR photography in which case this flaw is a boon

if your suggestion above is correct, the Olympus system would require a very specific IR cut filter to use it with older lenses; seems impractical (because of cost, mostly). At best (with an interference-based band filter) it would probably have similar problems as the B+W 486, i.e. strong hue shift towards the corner for all WA lenses.

exdeejjjaaaa said:

Rriley said:

the closest to this is Kodak's Truesence, which interpolates white

Click to expand...

those sensels in Oly patent are not "white" (where you do not have any color dye to separate spectrum), they are purely "IR" (all visible light is filtered out) and can't be used for anything other than CDAF assistance.

Click to expand...

Actually no, the patent shows them being Green during the image capture. As the CCD and CMOS sensels are very sensitive for Infrared, they do not need to have filter filtering visible light freqvencies. If you don't have IR cut filter in front of sensels, you get the Bayern filtered color AND IR in your signal. If you remove the IR filter from any CCD CMOS camera, you can take IR pictures with very little of additional stuff needed.

Basically what Oly seems to be doing is filtering the electronical signals from the sensels during the AF phase include only IR freq. In image capture phase, they take only the visible green light part of the signal. This could have been done so far already for all the colors instead of having IR filter, but having such filtering in all electronic paths would made it much more expensive than having simple optical filter.

I still don't understand then how they can record IR light on the image sensor (for the new AF system) AND block IR light for the normal imaging sensels at the same time (required because otherwise you get awful colors). I can't imagine that they have small IR blockers in front of all the normal color sensels, that sounds almost impossible, given the problems with IR blocking filters.

Or are you suggesting that they use the signal from the IR sensels to electronically remove IR contamination from neighbouring color sensels?

Posted in wrong place, ignore please

Blah brain fail

Rriley said:

Iforgetwhat8was4 said:

Anybody have a clue?

Click to expand...

yes, I posted it here
http://forums.dpreview.com/forums/read.asp?forum=1041&message=38773098

but it got lost among posts by this clueless moron who is trying to convince the world that 25% of the sensor is devoted to AF

Rriley said:

If CA magnification changes with focal length and not uniformly with different wavelengths, then you could use the difference in magnification between two colors as a range finder and get predictive AF.

But if you are just using the different focal points of different wavelengths you only get direction of focus and that wouldn't gain much...

Click to expand...

--
Riley

any similarity to persons living or dead is coincidental and unintended

Click to expand...

If the current focus point has visible light farther from focus and IR light closer to focus, well, I can see how the algorithm could determine how to push the visible back to focus. But if visible light and IR light are out of focus completly, how does the algorithm know which way to move the lens? Especially if all specturm of light are focused way behind the sensor or way in front of the sensor.

DjarumBlack said:

Rriley said:

Iforgetwhat8was4 said:

Anybody have a clue?

Click to expand...

yes, I posted it here
http://forums.dpreview.com/forums/read.asp?forum=1041&message=38773098

but it got lost among posts by this clueless moron who is trying to convince the world that 25% of the sensor is devoted to AF

DjarumBlack said:

If CA magnification changes with focal length and not uniformly with different wavelengths, then you could use the difference in magnification between two colors as a range finder and get predictive AF.

But if you are just using the different focal points of different wavelengths you only get direction of focus and that wouldn't gain much...

Click to expand...

--
Riley

any similarity to persons living or dead is coincidental and unintended

Click to expand...

If the current focus point has visible light farther from focus and IR light closer to focus, well, I can see how the algorithm could determine how to push the visible back to focus. But if visible light and IR light are out of focus completly, how does the algorithm know which way to move the lens? Especially if all specturm of light are focused way behind the sensor or way in front of the sensor.

Click to expand...

because one will still be out more than the other
its a bit like finding the smoothest bokeh, in reverse

although yes, i can recognise that eventually ultra fast lenses can be problematic like that

--
Riley

any similarity to persons living or dead is coincidental and unintended

technic said:

I still don't understand then how they can record IR light on the image sensor (for the new AF system) AND block IR light for the normal imaging sensels at the same time (required because otherwise you get awful colors). I can't imagine that they have small IR blockers in front of all the normal color sensels, that sounds almost impossible, given the problems with IR blocking filters.

Click to expand...

i dont see it as impossible, but certainly requiring precision
we dont know how its done, or if the bayer layer is intact

its possible, perhaps more likely that they just unshield half of the green pixels from IR light

technic said:

Or are you suggesting that they use the signal from the IR sensels to electronically remove IR contamination from neighbouring color sensels?

Click to expand...

--
Riley

any similarity to persons living or dead is coincidental and unintended

technic said:

Rriley said:

Thinking further, if they are manipulating the spectrum of IR light that comes through the lens, you might be able to duplicate that with an IR Cut filter screwed onto the lens for existing Zuikos, much the same way M8 operates. That would be intended to replace the coatings theyre using on the new lenses. I cant help thinking there has to be more to it than this though.

Click to expand...

why do you think Olympus is manipulating the IR spectrum with lens coatings? If IR doesn't come through the lens you cannot use it for this IR AF system. Or do you suggest the system uses e.g. the shorter IR wavelengths (like 700-900nm) for fast focusing, and removes the longer IR wavelengths because of adverse effects on the AF or colors in visible imaging?

Click to expand...

basically yes

technic said:

I don't think you can compare with the M8, because its problem is simply that its internal IR cut filter is too thin (weak), requiring an external IR cut for many normal lighting situations. The M8 doesn't need IR light at all to function, unless you use it for IR photography in which case this flaw is a boon

Click to expand...

the proposition was put, about using 4/3rds lenses on these mFT bodies

technic said:

if your suggestion above is correct, the Olympus system would require a very specific IR cut filter to use it with older lenses; seems impractical (because of cost, mostly). At best (with an interference-based band filter) it would probably have similar problems as the B+W 486, i.e. strong hue shift towards the corner for all WA lenses.

Click to expand...

nah not that expensive

--
Riley

any similarity to persons living or dead is coincidental and unintended

technic said:

I can't imagine that they have small IR blockers in front of all the normal color sensels, that sounds almost impossible, given the problems with IR blocking filters.

Click to expand...

what are the problems w/ IR blocking filters ? remember Olympus is an optical company - they have enough experience designing organic/inorganic materials that deal w/ light... so that is exactly where they have expertise probably better than their best friend Panasonic

--

This seems useful:

http://en.wikipedia.org/wiki/File:Achromatic_focal_curve.svg

and this, for PDAF:

http://www.scribd.com/doc/266790/Principle-of-the-Split-Image-Focusing-Aid-and-the-Phase-Comparison-Autofocus-Detector-in-Single-Lens-Reflex-Cameras

Nappe1 said:

Actually no, the patent shows them being Green during the image capture.

Click to expand...

even is so - that is not white

Nappe1 said:

As the CCD and CMOS sensels are very sensitive for Infrared, they do not need to have filter filtering visible light freqvencies.

Click to expand...

you filter out visible light so that your IR capture is not contaminated w/ portion of electrons generated by "green" photons - otherwise you have to try to calculate IR portion by using information from 4 neighboring pure "green" sensels... and that is approximation

Nappe1 said:

If you don't have IR cut filter in front of sensels, you get the Bayern filtered color AND IR in your signal. If you remove the IR filter from any CCD CMOS camera, you can take IR pictures with very little of additional stuff needed.

Click to expand...

thank you for the very valuable information, captain obvious.

Nappe1 said:

Basically what Oly seems to be doing is filtering the electronical signals from the sensels during the AF phase include only IR freq. In image capture phase, they take only the visible green light part of the signal.

Click to expand...

the question is - how do you know the ratio of IR light then vs green light that your IR sensel captured ? you can only approx guess that by using 4 regular (IR shielded) green sensels around that IR (where IR is not shielded) sensel , but then you do not really need that IR sensel to capture green light at all, as it is calculated anyways - so you can just leave that IR sensel as pure IR sensel (and have not appox calculation, but the exact IR capture... w/ noise of course), filter out all visible light and let firmware deall w/ that when the raw file is created to fake green by averaging 4 green IR shielded sensels... I can only imagine allowing green light through if it is economically cheaper to let green CFA dye in place, instead of putting something that will block all visible light completely... if it is cheaper than yes, they probably left green dye in place and then electronics does substract green based on the data from 4 green neighbors that are IR shielded to get IR signal and to get an image it again uses the data from 4 green neighbors that are IR shielded

Nappe1 said:

This could have been done so far already for all the colors instead of having IR filter

Click to expand...

the question is - how do you know the ratio of IR light then if all of your sensels are not IR shielded ?

--

Rriley said:

i dont see it as impossible, but certainly requiring precision
we dont know how its done, or if the bayer layer is intact

its possible, perhaps more likely that they just unshield half of the green pixels from IR light

Click to expand...

theoretically they could put the IR-cut layer (interference filter type, as in B+W 486) directly on the sensor manufacture the sensor so that the IR cut filter (probably a deposition process) will not stick to the IR sensels, or 'cut holes' for the IR sensels in the filter layer. But it sounds like a complicated job given the nature of the IR cut filter. And I guess this would rule out using these cameras for IR photography (unless you could get separate access to the IR sensels ...), because the IR cut filter would be permanently attached.

exdeejjjaaaa said:

what are the problems w/ IR blocking filters ? remember Olympus is an optical company - they have enough experience designing organic/inorganic materials that deal w/ light... so that is exactly where they have expertise probably better than their best friend Panasonic

Click to expand...

there are two approaches, absorption and interference filters. Problem is, the absorption filter need to be thick to have any effect (1-2 mm at least) which IMHO makes is unsuitable for just blocking IR from reaching the visible light sensels, assuming these are spread all over the chip. Even then you have a pretty gradual IR cut off which is not good for color accuracy. I guess it could be done if the IR sensels are clumped together in groups, but that is not my impression.

Most internal IR cut filters combine both absorption and interference filter for best result.

The interference filter is a layer of coatings (from what I remember over 20 layers) that remove the IR light with a certain wavelength. This could be build directly on the sensor, but as detailed in my post above this has other problems. I'm not saying it can't be done and I don't doubt Olympus expertise, it just sounds like a really complicated design with potential other issues (e.g. unsuitable for IR photography, although that is a really small market so they probably would not care anyway).

said otherwise, an IR cut filter is way more complicated than the Bayer filters used for the visible light sensels.

exdeejjjaaaa said:

exdeejjjaaaa said:

This could have been done so far already for all the colors instead of having IR filter

Click to expand...

the question is - how do you know the ratio of IR light then if all of your sensels are not IR shielded ?

Click to expand...

agree, if there is no real IR blocker for the color sensels, it seems to me color accuracy will be less good. There are various ways for estimating IR content in the signal, but it will only produce an estimate. I'm really curious what they are doing and how well it works.

Rriley said:

Rriley said:

if your suggestion above is correct, the Olympus system would require a very specific IR cut filter to use it with older lenses; seems impractical (because of cost, mostly). At best (with an interference-based band filter) it would probably have similar problems as the B+W 486, i.e. strong hue shift towards the corner for all WA lenses.

Click to expand...

nah not that expensive

Click to expand...

are you sure? the B+W 486 and similar filters are relatively expensive. Because of the hue shift I would prefer an absorption filter, but I don't know of any glass for filtering out a narrow near-IR band.