On 28/03/2023 19:24, John Larkin wrote:
> On Tue, 28 Mar 2023 07:05:16 -0700, Don Y
> <blockedofcourse@foo.invalid> wrote:
>
>> Stupid question: do different wavelengths "focus" at different
>> depths? I know different colors of visible light do (orange vs blue?)
>> but assumed that was a mechanical consequence of the eye's structure...
>> I.e., as there is one lens/aperture, does it need to be tweeked
>> when switching from visible light mode to Ir mode?
>
> Achromatic lenses were invented to focus all wavelengths the same. I
> wonder if a human eys is achromatic.

Achromatic lenses focus two specific wavelengths of light to the same
point. Usually blue 495nm and orange wavelength 590nm (Na).

APOchromatic lenses focus three specific wavelengths to the same point
and additionally are corrected for spherical aberration at two
wavelengths rather than just one. They still have a tiny amount of
residual colour on demanding high dynamic range targets.

https://en.wikipedia.org/wiki/Apochromat

Some lens makers are less than honest about their APO designation.

--
Martin Brown

On Wed, 29 Mar 2023 18:22:57 +0100, Martin Brown
<'''newspam'''@nonad.co.uk> wrote:

>On 28/03/2023 19:24, John Larkin wrote:
>> On Tue, 28 Mar 2023 07:05:16 -0700, Don Y
>> <blockedofcourse@foo.invalid> wrote:
>>
>>> Stupid question: do different wavelengths "focus" at different
>>> depths? I know different colors of visible light do (orange vs blue?)
>>> but assumed that was a mechanical consequence of the eye's structure...
>>> I.e., as there is one lens/aperture, does it need to be tweeked
>>> when switching from visible light mode to Ir mode?
>>
>> Achromatic lenses were invented to focus all wavelengths the same. I
>> wonder if a human eys is achromatic.
>
>Achromatic lenses focus two specific wavelengths of light to the same
>point. Usually blue 495nm and orange wavelength 590nm (Na).
>
>APOchromatic lenses focus three specific wavelengths to the same point
>and additionally are corrected for spherical aberration at two
>wavelengths rather than just one. They still have a tiny amount of
>residual colour on demanding high dynamic range targets.
>
>https://en.wikipedia.org/wiki/Apochromat
>
>Some lens makers are less than honest about their APO designation.

We have two eyes and each can dither its focal length, and our brains
do amazing processing to construct 3D images. So we can compute around
all sorts of distortions, not just different focal lengths vs color.

On 2023-03-29 13:22, Martin Brown wrote:
> On 28/03/2023 19:24, John Larkin wrote:
>> On Tue, 28 Mar 2023 07:05:16 -0700, Don Y
>> <blockedofcourse@foo.invalid> wrote:
>>
>>> Stupid question:  do different wavelengths "focus" at different
>>> depths?  I know different colors of visible light do (orange vs blue?)
>>> but assumed that was a mechanical consequence of the eye's structure...
>>> I.e., as there is one lens/aperture, does it need to be tweeked
>>> when switching from visible light mode to Ir mode?
>>
>> Achromatic lenses were invented to focus all wavelengths the same. I
>> wonder if a human eys is achromatic.
>
> Achromatic lenses focus two specific wavelengths of light to the same
> point. Usually blue 495nm and orange wavelength 590nm (Na).
>
> APOchromatic lenses focus three specific wavelengths to the same point
> and additionally are corrected for spherical aberration at two
> wavelengths rather than just one. They still have a tiny amount of
> residual colour on demanding high dynamic range targets.
>
> https://en.wikipedia.org/wiki/Apochromat
>
> Some lens makers are less than honest about their APO designation.
>

Nobody designs lenses like that anymore, though. I'm not a lens
designer myself, but I'm told that you just pick a wavelength interval
and FOV, then tell the lens design package to minimize the RMS ray
intercept error.

"Apochromat", "Aplanat" and such like nowadays just mean
"German-expensive-nice". ;)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

On 3/29/2023 9:14 AM, Martin Brown wrote:
>> Overriding the existing (light level based?) control of the filter
>> would mean I would experience the rolloff of color response to lower
>> light levels (?).  I.e., I could see just how "badly" the camera handles
>> color in those cases.
>>
>> And, how "impure" the color response would be at higher light levels (?).
>
> That may well vary a fair bit from camera to camera depending on their exact
> choice of RGB Bayer filters (other permutations exist).

Manufacturer to manufacturer? Model to model? Or, unit to unit?

> I found that on pure H-alpha red light at 656.3nm leaked through the green
> filter enough to produce a crazy polka dot yellow and red pattern on one of my
> early one shot colour camera. The video camera was fine with it although the
> photosphere washed out but some single shot cameras really didn't get on with
> imaging in monochromatic red light at all.

Hmmm... I can't say I've ever had opportunity/need to do so. Though Ir
would be closest.

>> But, is the sensor "happier" working in Ir?  I.e., if I replaced the
>> Ir emitters with an "equivalent" amount of visible light, would I
>> retain the detail that is evident under Ir illumination AND get the
>> benefit of color response at those lower ENVIRONMENTAL light levels?
>>
>> Said another way, why were Ir illuminators chosen instead of something
>> that emits visible light?  Is it so your camera is NOT acting as a (visible)
>> spotlight?  Or, the relative cost of solid state visible emitters?  Or,
>> some characteristic of the sensor?
>
> Matching to the sensor wavelength response. Optimum would be red 600nm
> nominally 100% but 90% at 700nm or 60% AT 800nm isn't bad. Raw sensitivity
> curve is on p10 of this datasheet:
>
> https://s1-dl.theimagingsource.com/api/2.5/packages/publications/sensor-sony/icx618ala/cfcc5312-8e4f-5524-a5fb-9bcb0f728ca8/icx618ala_1.2.en_US.pdf
>>
>> Note that many newer cameras now are accompanied by (visible light) spotlights.
>> It seems like the best option would be to use Ir emitters in low light levels
>> "for their invisibility" so that motion detection could be performed using
>> just the luminance information available.  *Then*, activate a visible light
>> source if you wanted to capture more detail and/or color...
>
> I think that is the main reason. I can see some of the ones with 750nm
> illuminators - they have a tail that is just into visible red for my eyes. I
> can't see 800nm at all. I have a set of low wave filters.

I can see the emitters glowing red (if I look directly into them). But, can't
see anything cast from them.

They work well at illuminating the (indoor) scenes that I've tried.
But, those are all short range (distance) targets; ceiling to floor
(about 10-12 ft) and across the width of the garage door (~19 ft).
I haven't tried them out-of-doors. (I need to install a few to watch
the top of the roof which will be ~30 ft if I site them on opposite
ends of the house and hope for a little overlap in the middle).

This is all a learning experience; figure out what you can do with X
and then figure out if you need to do more or can afford to do less!

On 3/29/2023 9:33 AM, Martin Brown wrote:
> On 28/03/2023 19:55, Don Y wrote:
>
>> Said in more practical terms:  when the camera switches to
>> Ir mode, are any other changes made to the optics to
>> compensate for this different illumination wavelength?
>
> It might shift it slightly to compensate (or removing the IR blocking filter
> from the optical path might have the same effect). Smaller lenses tend to have
> a better depth of field so it is less pronounced.
>
>> You have two "similar" cameras.  One takes color photos and the
>> other takes B&W.  If you are not interested in the chroma channel
>> (e.g., maybe you're colorblind... or, a computer algorithm that
>> is going to deliberately discard chroma), is there a difference
>> in the quality/detail/whatever of the two images?
>
> OK this is answerable. If you have a pure monochrome camera
>
> You get a matrix of pixel data all with weight 1 so that any 2x2 block contains
> exactly 4 distinct independent observations of the luminance.
>
> Y Y    as    1 1
> Y Y        1 1
>
> In a classic Bayer mask you get green, red and blue subsampled and the
> luminance Y is later interpolated from these data.
>
> G R    as    0.59 0.30
> B G        0.11 0.59
>
> SO you have thrown away a factor of two in resolution on R & B and a factor of
> root2 on the green channel (and twisted the sampling axis). The total luminance
> captured is 1.59 instead of 4 so your signal on the luminance channel is on
> average 0.4 instead of 1.

Ah, OK. Does this mask repeat as:

G R G R G R
B G B G B G
G R G R...

or, perhaps

G R G B G R G B
B G R G B G R G
G B G R...

(i.e., more G than R or B)

E.g., an RGB(W) illuminator has equal amounts of each emitter.

> This will bite you in the backside when photographing objects in red light like
> H-alpha or red flowers with dark veins on them like poppies.
> Right shade of blue flowers are rarers but tree branches against a string blue
> sky can also end up mutilated by Bayer demosaic. The algorithms are much
> smarter today but with the right/wrong sort of designer  target image there are
> invariably a few pathological cases.

But, what feature level do you need to buy at to get the advantage of
these algorithms? E.g., do webcams have the algorithms embedded?
Or, in the hosted driver?

I.e, what do *I* have to do given a particular choice of image sensor
(or device)?

On 29/03/2023 21:44, Don Y wrote:
> On 3/29/2023 9:33 AM, Martin Brown wrote:
>> On 28/03/2023 19:55, Don Y wrote:
>>
>>> Said in more practical terms:  when the camera switches to
>>> Ir mode, are any other changes made to the optics to
>>> compensate for this different illumination wavelength?
>>
>> It might shift it slightly to compensate (or removing the IR blocking
>> filter from the optical path might have the same effect). Smaller
>> lenses tend to have a better depth of field so it is less pronounced.
>>
>>> You have two "similar" cameras.  One takes color photos and the
>>> other takes B&W.  If you are not interested in the chroma channel
>>> (e.g., maybe you're colorblind... or, a computer algorithm that
>>> is going to deliberately discard chroma), is there a difference
>>> in the quality/detail/whatever of the two images?
>>
>> OK this is answerable. If you have a pure monochrome camera
>>
>> You get a matrix of pixel data all with weight 1 so that any 2x2 block
>> contains exactly 4 distinct independent observations of the luminance.
>>
>> Y Y    as    1 1
>> Y Y        1 1
>>
>> In a classic Bayer mask you get green, red and blue subsampled and the
>> luminance Y is later interpolated from these data.
>>
>> G R    as    0.59 0.30
>> B G        0.11 0.59
>>
>> SO you have thrown away a factor of two in resolution on R & B and a
>> factor of root2 on the green channel (and twisted the sampling axis).
>> The total luminance captured is 1.59 instead of 4 so your signal on
>> the luminance channel is on average 0.4 instead of 1.
>
> Ah, OK.  Does this mask repeat as:
>
> G R G R G R
> B G B G B G
> G R G R...
>
> or, perhaps
>
> G R G B G R G B
> B G R G B G R G
> G B G R...
>
> (i.e., more G than R or B)

Yes Chroma subsampled once per 2x2 block. However, some cameras use CMYG
or two shades of green to get a wider gamut and a bit more sensitivity.

Depending on what you want to do it is worth cannibalising a normal
webcam to try things out. QUIAG has various recipes for doing ultra low
light level work with specific webcams with unusually low noise sensors.

It is an amateur astronomy group but some of the guys there have
considerable experience in modifying off the shelf kit to ultra low
light sensitivity way beyond what the manufacturers had ever intended.
>
> E.g., an RGB(W) illuminator has equal amounts of each emitter.
>
>> This will bite you in the backside when photographing objects in red
>> light like H-alpha or red flowers with dark veins on them like poppies.
>> Right shade of blue flowers are rarers but tree branches against a
>> string blue sky can also end up mutilated by Bayer demosaic. The
>> algorithms are much smarter today but with the right/wrong sort of
>> designer  target image there are invariably a few pathological cases.
>
> But, what feature level do you need to buy at to get the advantage of
> these algorithms?  E.g., do webcams have the algorithms embedded?
> Or, in the hosted driver?

Back when I first got my first digital camera, a Kodak DC-120 I did some
experiments on its raw data mode. Last modified date is misleading and
caused by transfer of site after my ISP went pop. It was about 1998.

http://nezumidemon.co.uk/photo/bayer/bayer.html

This isn't a bad introduction to the nitty gritty demosaic formulae:

https://web.archive.org/web/20230201063227/http://www.siliconimaging.com:80/RGB%20Bayer.htm

These days I don't know but it used to be the case that the hardware was
basically a readout device and a frame buffer and all the smarts were in
the PC. There are hardware realtime MPEG encoders now that are probably
internal to the camera so that it spits out an MPEG stream on USB.
>
> I.e, what do *I* have to do given a particular choice of image sensor
> (or device)?

Potentially quite a lot which is why I'd suggest finding an off the peg
webcam that is close enough for what you want and stick with it.

Depending on how low light you want to go these guys have some
interesting tricks and hacks specific to particular models of cheap (and
not so cheap) webcam. Some are a lot more adjustable than others.

http://www.qcuiag.org.uk

--
Martin Brown

On 3/29/2023 2:37 PM, Martin Brown wrote:
>> G R G R G R
>> B G B G B G
>> G R G R...
>>
>> or, perhaps
>>
>> G R G B G R G B
>> B G R G B G R G
>> G B G R...
>>
>> (i.e., more G than R or B)
>
> Yes Chroma subsampled once per 2x2 block. However, some cameras use CMYG or two
> shades of green to get a wider gamut and a bit more sensitivity.
>
> Depending on what you want to do it is worth cannibalising a normal webcam to
> try things out.

I'm having a hard time picking an implementation approach.
There are "composite video" cameras, USB cameras, bare sensors,
etc.

Then, things like PTZ, Ir illuminators, various sizes and
form factors, etc.

I've been trying to adopt a single "camera interface" that I
can deploy in different configurations/applications. That
is proving ... challenging!

E.g., I use cameras as sensors (is anything in the path of the
closing/opening garage door? has the mailman visited the mailbox?
is the sky overcast vs. clear?), for surveillance/security,
to track occupants -- and visitors (*who* is at the front door?), etc.

The markets for various devices all seem to approach the
problem differently. E.g., I'm more likely to find a PTZ
camera with a network connection *or* composite video out
than a USB device. OTOH, a pinhole/"spy" camera would more
likely be composite video or USB.

And, USB seems to have a lot higher turnover (models) as
it directly feeds the consumer/PC market and last year's
model may no longer be available THIS year (not a good strategy
for a product with along lifespan).

I take a hint from folks designing cameras for these varying
markets: you don't see them building a product around a
COTS web cam as the core "image capture module" -- despite
the fact that there is so much price pressure on those
offerings! You can *make* a dashcam with a webcam... but
if you buy one, it most likely has integrated *a* camera
instead of building on some other offering.

> QUIAG has various recipes for doing ultra low light level work
> with specific webcams with unusually low noise sensors.
>
> It is an amateur astronomy group but some of the guys there have considerable
> experience in modifying off the shelf kit to ultra low light sensitivity way
> beyond what the manufacturers had ever intended.

I will have a look. But, in light of the above, suspect it will
just be educational and not directly applicable.

>> E.g., an RGB(W) illuminator has equal amounts of each emitter.
>>
>>> This will bite you in the backside when photographing objects in red light
>>> like H-alpha or red flowers with dark veins on them like poppies.
>>> Right shade of blue flowers are rarers but tree branches against a string
>>> blue sky can also end up mutilated by Bayer demosaic. The algorithms are
>>> much smarter today but with the right/wrong sort of designer  target image
>>> there are invariably a few pathological cases.
>>
>> But, what feature level do you need to buy at to get the advantage of
>> these algorithms?  E.g., do webcams have the algorithms embedded?
>> Or, in the hosted driver?
>
> Back when I first got my first digital camera, a Kodak DC-120 I did some
> experiments on its raw data mode. Last modified date is misleading and caused
> by transfer of site after my ISP went pop. It was about 1998.
>
> http://nezumidemon.co.uk/photo/bayer/bayer.html

This is happening *in* the camera (as the resulting JPG -- or whatever -- has
to be "consumable" by the user)?

> This isn't a bad introduction to the nitty gritty demosaic formulae:
>
> https://web.archive.org/web/20230201063227/http://www.siliconimaging.com:80/RGB%20Bayer.htm
>
> These days I don't know but it used to be the case that the hardware was
> basically a readout device and a frame buffer and all the smarts were in the
> PC. There are hardware realtime MPEG encoders now that are probably internal to
> the camera so that it spits out an MPEG stream on USB.

Yeah, I think this has become so commonplace that it "made sense"
to build silicon for the job.

And, because the camera is a "defined functionality" product; it
doesn't have to be able to process data from a different sensor and
deliver it in the /encoding du jour/.

>> I.e, what do *I* have to do given a particular choice of image sensor
>> (or device)?
>
> Potentially quite a lot which is why I'd suggest finding an off the peg webcam
> that is close enough for what you want and stick with it.

But "sticking with it" for more than a year (two) is likely
not possible. You'd have to do a lifetime buy just to be
assured an ongoing supply of the device.

I'm not sure how much better (worse?) the situation is if you
go the bare sensor approach.

> Depending on how low light you want to go these guys have some interesting
> tricks and hacks specific to particular models of cheap (and not so cheap)
> webcam. Some are a lot more adjustable than others.
>
> http://www.qcuiag.org.uk

I'm not particularly driven by "low light" in the extreme sense.
Rather, just trying to understand what I can get for modest
effort/cost.

E.g., I *could* switch on the overhead lights in the garage if
I wanted to see what's in that arena in what would otherwise
be "low light" conditions. Or, augment a security camera
with Ir (or visible light) illuminators. Ditto for a door cam.

But, if I don't have to do so, then its folly to add that requirement!

Thanks, I'll have a look at the URLs!

On 29/03/2023 21:29, Phil Hobbs wrote:
> On 2023-03-29 13:22, Martin Brown wrote:
>> On 28/03/2023 19:24, John Larkin wrote:
>>> On Tue, 28 Mar 2023 07:05:16 -0700, Don Y
>>> <blockedofcourse@foo.invalid> wrote:
>>>
>>>> Stupid question:  do different wavelengths "focus" at different
>>>> depths?  I know different colors of visible light do (orange vs blue?)
>>>> but assumed that was a mechanical consequence of the eye's structure...
>>>> I.e., as there is one lens/aperture, does it need to be tweeked
>>>> when switching from visible light mode to Ir mode?
>>>
>>> Achromatic lenses were invented to focus all wavelengths the same. I
>>> wonder if a human eys is achromatic.
>>
>> Achromatic lenses focus two specific wavelengths of light to the same
>> point. Usually blue 495nm and orange wavelength 590nm (Na).
>>
>> APOchromatic lenses focus three specific wavelengths to the same point
>> and additionally are corrected for spherical aberration at two
>> wavelengths rather than just one. They still have a tiny amount of
>> residual colour on demanding high dynamic range targets.
>>
>> https://en.wikipedia.org/wiki/Apochromat
>>
>> Some lens makers are less than honest about their APO designation.
>>
>
> Nobody designs lenses like that anymore, though.  I'm not a lens
> designer myself, but I'm told that you just pick a wavelength interval
> and FOV, then tell the lens design package to minimize the RMS ray
> intercept error.

I don't doubt it. The other thing that has happened in digital is that
demosaicing algorithms remove the previous classic film restriction that
there should be no lateral chromatic aberration. So long as each R,G,B
image is in focus on the same sensor plane clever demosaicing can sort
out the slightly different magnifications and certain aberrations too.

The optics now only have to do all the things that software can't do.

There was a time when a 3x optical zoom was about the limit for a cheap
camera but now there are some with 10x (or more).

> "Apochromat", "Aplanat" and such like nowadays just mean
> "German-expensive-nice". ;)

Or Japanese - in the APO telescopes line Takahashi is hard to beat.
There are some even more esoteric Japanese telescope makers too.

--
Martin Brown