Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Hasselblad's answer to the Foveon

1 view
Skip to first unread message

Alan Browne

unread,
Nov 22, 2009, 2:28:05 PM11/22/09
to
http://www.dpreview.com/news/0911/09111902hasselbladh3d50ms.asp

For still photography, every pixel is shot with each colour filter by
mechanically moving the filter array by 1 pixel between shots. 4 shots
later you have 50 Mpix with true RGB for each pixel. At that, it is 16
bits/colour. No demosaicing means more accurate colour at each pixel
and sharper images.

For repros (of paintings), product, etc. this will be phenomenal.

(At 1.4s per capture, an image will take nearly 6 seconds to take...
better have a very sturdy tripod).

The camera can also be used in normal mosaic mode.

RichA

unread,
Nov 22, 2009, 6:04:01 PM11/22/09
to

Any tripod that can hold a camera steady from shot to shot with less
than say a 1/4 micron shift would have to be one hell of a tripod.
http://www.pbase.com/andersonrm/image/101130219

Message has been deleted

Alan Browne

unread,
Nov 22, 2009, 8:48:56 PM11/22/09
to
John A. wrote:
> I figure any tripod good enough for that many megapixels and a 6
> second single exposure can do the job. It's not like you're going to
> be using the shutter release button on the camera body, after all.
>
> Plus, the shift tolerance would be greater than that, and vary with
> the length of the lens. It would be particularly high for wide angle
> shots. Remember: it's the amount of shift needed to make a part of the
> scene shift one pixel, not the width of the pixel on the sensor.

If I understand it correctly, the entire filter array is moved
vertically or horizontally by 1 pixel height/width for each of the 4
shots. See the Hasselblad site. There is a graphic.

Message has been deleted

Alan Browne

unread,
Nov 23, 2009, 3:45:44 PM11/23/09
to
John A. wrote:
> Exactly. Rich's concern is that the camera might shift between shots,
> throwing off the registration.

Rich can go look at the sample shots at the Hasselblad site if he's not
convinced.

It's funny, somehow, that I trust Hasselblad more than RichA.

I must be a cynic.

Alan Browne

unread,
Nov 23, 2009, 4:07:57 PM11/23/09
to
John A. wrote:
> On Sun, 22 Nov 2009 20:48:56 -0500, Alan Browne
> <alan....@FreelunchVideotron.ca> wrote:
>
> Exactly. Rich's concern is that the camera might shift between shots,
> throwing off the registration.

I also meant to state that the filter array, indeed, has to shift by a
whole pixel size - that is to say move the center of one pixel to the
center of the next. I'd expect the position tolerance to be on the
order of less than 5 - 10% (and that there is sufficient anti-leak masking).

John A.

unread,
Nov 23, 2009, 5:08:49 PM11/23/09
to
On Mon, 23 Nov 2009 16:07:57 -0500, Alan Browne
<alan....@FreelunchVideotron.ca> wrote:

From what I gather, it shifts the whole sensor, not just the filter.

John A.

unread,
Nov 23, 2009, 5:23:38 PM11/23/09
to
On Mon, 23 Nov 2009 17:08:49 -0500, John A. <jo...@nowhere.invalid>
wrote:

Come to think of it, IIRC I suggested something along these lines a
while back. Gotta look that up. When did H come out with the first
one?

Alan Browne

unread,
Nov 23, 2009, 5:30:17 PM11/23/09
to

Whichever. One relative to the other in any case.

Alan Browne

unread,
Nov 23, 2009, 5:39:13 PM11/23/09
to

No idea.

I did think of having a full over separate R then G then B filter that
were mechanically moved over the sensor (but you lose the RGB array) and
then you take 3 shots in succession. With a 35mm sized sensor you could
probably do this in about 25 ms per shot - much faster than the shoot rate.

It might be worth taking an old 6 mpix camera and seeing if the RGB
filter can be removed along with the AA filter. But I bet the RGB
filter is under the "window" on the sensor chip and glued in place over
the sensor itself.

It would take a little s/w work (post process) but dcraw could be hacked
for the purpose.

John A.

unread,
Nov 23, 2009, 5:51:59 PM11/23/09
to
On Mon, 23 Nov 2009 17:30:17 -0500, Alan Browne
<alan....@FreelunchVideotron.ca> wrote:

No, the sensor+filter array all together. You lose a row and a column,
but what the hey?

nospam

unread,
Nov 23, 2009, 6:25:47 PM11/23/09
to
In article <5qKdnROjn6OMkpbW...@giganews.com>, Alan Browne
<alan....@FreelunchVideotron.ca> wrote:

> It might be worth taking an old 6 mpix camera and seeing if the RGB
> filter can be removed along with the AA filter. But I bet the RGB
> filter is under the "window" on the sensor chip and glued in place over
> the sensor itself.

someone on dpreview has done exactly that by bleaching off the colour
filters. i think he may have written custom raw processing code as
well.

Alan Browne

unread,
Nov 24, 2009, 9:21:42 AM11/24/09
to

Not sure what you mean. If they both move, you get nothing.

Alan Browne

unread,
Nov 24, 2009, 9:22:26 AM11/24/09
to

Interesting, I'll look into it.

John A.

unread,
Nov 24, 2009, 10:09:06 AM11/24/09
to
On Tue, 24 Nov 2009 09:21:42 -0500, Alan Browne
<alan....@FreelunchVideotron.ca> wrote:

You get the image projected by the lens shifted one pixel each shot.
(Or rather the sensor shifted around in the projected light.) So each
pixel-sized spot of light is captured in turn by different colored
bayer-array sensors.

Alan Browne

unread,
Nov 24, 2009, 10:16:04 AM11/24/09
to

You're not explaining anything clearly. There is no need to move both.

The point of the exercise is to expose each sensor to a different color
filtered light at least once. By moving the colour filter array over
the sensor -or- moving the sensor under the filter array that is
achieved. There is no need to move both, and it seems to me most
logical to move only the colour filter array as it is presumably
lighter. However, per the Hasselblad brochure:

"High precision piezo motors control movements of the sensor in one
pixel increments. By combining four shots, each offset by one pixel,
the true colours, Red, Green and Blue of each point are obtained."

So, only one is moved and it is the sensor. For registration purposes
this means the measured array needs to be shifted numerically to
register properly.

David J. Littleboy

unread,
Nov 24, 2009, 10:40:07 AM11/24/09
to

"Alan Browne" <alan....@FreelunchVideotron.ca> wrote:
>>
>> You get the image projected by the lens shifted one pixel each shot.
>> (Or rather the sensor shifted around in the projected light.) So each
>> pixel-sized spot of light is captured in turn by different colored
>> bayer-array sensors.
>
> You're not explaining anything clearly. There is no need to move both.

Sure there is: it's physically impossible to move the filter array
independently of the sensor.

> The point of the exercise is to expose each sensor to a different color
> filtered light at least once.

No, it's not.

The point is to use a _different_ physical pixel to record each of R, G, and
B at each _location_ in the image plane.

Then they reconstruct each image plane pixel from RGB measurements made with
different physical pixels, that happened to be at the same place (at
different times) when they made their measurments.

> By moving the colour filter array over the sensor -or- moving the sensor
> under the filter array that is achieved. There is no need to move both,
> and it seems to me most logical to move only the colour filter array as it
> is presumably lighter.

Everything I've ever read about Bayer sensors has the color filters
fabricated on the chip under the microlenses (which are also fabricated on
the chip), so there's no way to move the filter array independently of the
sensor.

And evern if it were possible, it'd be really silly to do that instead of
just using in-camera IS technology to move the sensor.

>However, per the Hasselblad brochure:
>
> "High precision piezo motors control movements of the sensor in one
> pixel increments. By combining four shots, each offset by one pixel,
> the true colours, Red, Green and Blue of each point are obtained."
>
> So, only one is moved and it is the sensor. For registration purposes
> this means the measured array needs to be shifted numerically to register
> properly.

It only makes sense to move the whole sensor assembly, as is done with
in-camera image stabilization. (I.e. there's nothing "new" here at all: it's
just using an existing technology to do something different.)

--
David J. Littleboy
Tokyo, Japan


Alan Browne

unread,
Nov 24, 2009, 12:24:49 PM11/24/09
to
David J. Littleboy wrote:
> "Alan Browne" <alan....@FreelunchVideotron.ca> wrote:
>>> You get the image projected by the lens shifted one pixel each shot.
>>> (Or rather the sensor shifted around in the projected light.) So each
>>> pixel-sized spot of light is captured in turn by different colored
>>> bayer-array sensors.
>> You're not explaining anything clearly. There is no need to move both.
>
> Sure there is: it's physically impossible to move the filter array
> independently of the sensor.
>
>> The point of the exercise is to expose each sensor to a different color
>> filtered light at least once.
>
> No, it's not.
>
> The point is to use a _different_ physical pixel to record each of R, G, and
> B at each _location_ in the image plane.

You better go read the Hasselblad brochure, for that is precisely what
they are doing: moving the sensor wrt to the filter array. That it
takes nearly 6 seconds to take one photo (4 images) is another matter.
(This has to do with the very slow frame rate of the high Mpix Hassy's,
about 1.6 s per image).


>
> Then they reconstruct each image plane pixel from RGB measurements made with
> different physical pixels, that happened to be at the same place (at
> different times) when they made their measurments.
>
>> By moving the colour filter array over the sensor -or- moving the sensor
>> under the filter array that is achieved. There is no need to move both,
>> and it seems to me most logical to move only the colour filter array as it
>> is presumably lighter.
>
> Everything I've ever read about Bayer sensors has the color filters
> fabricated on the chip under the microlenses (which are also fabricated on
> the chip), so there's no way to move the filter array independently of the
> sensor.

Again, pls see the Hassy literature.

>
> And evern if it were possible, it'd be really silly to do that instead of
> just using in-camera IS technology to move the sensor.

Again, pls see the Hassy literature.

>
>> However, per the Hasselblad brochure:
>>
>> "High precision piezo motors control movements of the sensor in one
>> pixel increments. By combining four shots, each offset by one pixel,
>> the true colours, Red, Green and Blue of each point are obtained."
>>
>> So, only one is moved and it is the sensor. For registration purposes
>> this means the measured array needs to be shifted numerically to register
>> properly.
>
> It only makes sense to move the whole sensor assembly, as is done with
> in-camera image stabilization. (I.e. there's nothing "new" here at all: it's
> just using an existing technology to do something different.)

Well, it is new because no other camera manufacturer is doing it.

Pls see the Hassy literature.

John A.

unread,
Nov 24, 2009, 12:42:12 PM11/24/09
to
On Wed, 25 Nov 2009 00:40:07 +0900, "David J. Littleboy"
<dav...@gol.com> wrote:

>
>"Alan Browne" <alan....@FreelunchVideotron.ca> wrote:
>>>
>>> You get the image projected by the lens shifted one pixel each shot.
>>> (Or rather the sensor shifted around in the projected light.) So each
>>> pixel-sized spot of light is captured in turn by different colored
>>> bayer-array sensors.
>>
>> You're not explaining anything clearly. There is no need to move both.
>
>Sure there is: it's physically impossible to move the filter array
>independently of the sensor.

At least not without radically redesigning the filter+sensor assembly,
and introducing an air gap or abrasion between the two. More trouble
than it's worth. Wisely, H has seen that just moving the whole shebang
is simpler and just as effective.

>> The point of the exercise is to expose each sensor to a different color
>> filtered light at least once.
>
>No, it's not.
>
>The point is to use a _different_ physical pixel to record each of R, G, and
>B at each _location_ in the image plane.

Actually the point is to have each location in the image captured by
each of the colors in turn. Using a different physical pixel to do so
is just the way H is doing it, and seems a natural way to do it given
a Bayer array. But I know that's what you meant. ;)

There are other ways, of course, such as the Foveon layered sensor, or
using three full image- or lens- sized filters with an otherwise
unfiltered sensor. There are tradeoffs for each of them.

>Then they reconstruct each image plane pixel from RGB measurements made with
>different physical pixels, that happened to be at the same place (at
>different times) when they made their measurments.

Yep. I think Alan's having a mental block about how the final image is
assembled from the exposures.

>> By moving the colour filter array over the sensor -or- moving the sensor
>> under the filter array that is achieved. There is no need to move both,
>> and it seems to me most logical to move only the colour filter array as it
>> is presumably lighter.
>
>Everything I've ever read about Bayer sensors has the color filters
>fabricated on the chip under the microlenses (which are also fabricated on
>the chip), so there's no way to move the filter array independently of the
>sensor.
>
>And evern if it were possible, it'd be really silly to do that instead of
>just using in-camera IS technology to move the sensor.

Exactly.

>>However, per the Hasselblad brochure:
>>
>> "High precision piezo motors control movements of the sensor in one
>> pixel increments. By combining four shots, each offset by one pixel,
>> the true colours, Red, Green and Blue of each point are obtained."
>>
>> So, only one is moved and it is the sensor. For registration purposes
>> this means the measured array needs to be shifted numerically to register
>> properly.
>
>It only makes sense to move the whole sensor assembly, as is done with
>in-camera image stabilization. (I.e. there's nothing "new" here at all: it's
>just using an existing technology to do something different.)

Just so. Makes me wonder what the precision is on existing DSLR
sensor-based IS actuators, and if any of the cameras out there could
possibly achieve this with just a firmware upgrade.

John A.

unread,
Nov 24, 2009, 1:02:51 PM11/24/09
to
On Tue, 24 Nov 2009 12:24:49 -0500, Alan Browne
<alan....@FreelunchVideotron.ca> wrote:

>David J. Littleboy wrote:
>> "Alan Browne" <alan....@FreelunchVideotron.ca> wrote:
>>>> You get the image projected by the lens shifted one pixel each shot.
>>>> (Or rather the sensor shifted around in the projected light.) So each
>>>> pixel-sized spot of light is captured in turn by different colored
>>>> bayer-array sensors.
>>> You're not explaining anything clearly. There is no need to move both.
>>
>> Sure there is: it's physically impossible to move the filter array
>> independently of the sensor.
>>
>>> The point of the exercise is to expose each sensor to a different color
>>> filtered light at least once.
>>
>> No, it's not.
>>
>> The point is to use a _different_ physical pixel to record each of R, G, and
>> B at each _location_ in the image plane.
>
>You better go read the Hasselblad brochure, for that is precisely what
>they are doing: moving the sensor wrt to the filter array. That it
>takes nearly 6 seconds to take one photo (4 images) is another matter.
>(This has to do with the very slow frame rate of the high Mpix Hassy's,
>about 1.6 s per image).

I've seen it. They're moving the whole array. Otherwise each spot in
the image would still be captured with a Bayer array pattern that does
not achieve the desired effect.

The filters *have to* move for this scheme to work. Moving the sensor
but not the filters does nothing to achieve the desired effect of
R+G+B capture at every pixel. They *are* moving the sensor so that
*must* include the filters.

Also it would be needlessly complicated on an engineering level to do
it any other way. By moving the whole sensor+filter assembly they
avoid a radical change in sensor+filter fabrication, sensor+filter
registration issues, physical wear, etc. They might use existing IS
hardware if that's precise enough. They might add special actuators
designed to simply and precisely move the sensor one exact pixel
width/height at a time. Either way they move it, it works, and very
simply.

Mr. Commentary

unread,
Nov 24, 2009, 1:05:58 PM11/24/09
to

I would have preferred to just answer with, "Oy, you're a blooming idiot,
again." And leave it at that. Maybe he would have tried to exercise one or
two if his troll's remaining brain-cells.

Mr. Commentary

unread,
Nov 24, 2009, 1:08:11 PM11/24/09
to
On Tue, 24 Nov 2009 10:16:04 -0500, Alan Browne
<alan....@FreelunchVideotron.ca> wrote:

Hah! I see even explaining it to him didn't help. Perhaps he's all out of
troll's brain-cells.

John A.

unread,
Nov 24, 2009, 1:27:18 PM11/24/09
to

You used them both up. Now go rest your little head.

Alan Browne

unread,
Nov 24, 2009, 2:41:48 PM11/24/09
to

You don't seem to get that _relative_ to one another, one is moving,
otherwise it's damned hard to get a different color over a pixel location.

>
> Also it would be needlessly complicated on an engineering level to do
> it any other way. By moving the whole sensor+filter assembly they
> avoid a radical change in sensor+filter fabrication, sensor+filter
> registration issues, physical wear, etc. They might use existing IS
> hardware if that's precise enough. They might add special actuators
> designed to simply and precisely move the sensor one exact pixel
> width/height at a time. Either way they move it, it works, and very
> simply.

They use piezoelectric actuators. Such can be made to move by very tiny
(and accurate) movements (as is obviously required here).

David J. Littleboy

unread,
Nov 24, 2009, 3:31:53 PM11/24/09
to

"Alan Browne" <alan....@FreelunchVideotron.ca> wrote:
> John A. wrote:
>>
>> I've seen it. They're moving the whole array. Otherwise each spot in
>> the image would still be captured with a Bayer array pattern that does
>> not achieve the desired effect.
>>
>> The filters *have to* move for this scheme to work. Moving the sensor
>> but not the filters does nothing to achieve the desired effect of
>> R+G+B capture at every pixel. They *are* moving the sensor so that
>> *must* include the filters.
>
> You don't seem to get that _relative_ to one another, one is moving,
> otherwise it's damned hard to get a different color over a pixel location.

It's easy: they just shift the data. Assume pixel 759,759 in the sensor is
R. Shift the whole sensor shebang one pixel position left, and now pixel
759,760 in the sensor (B) is measuring the same point in the image. Shift
vertically, and pixel 760,760 in the sensor (G) is measuring the same point
in the image.

Now construct pixel 759,759 in the data from pixels 759,759, 759,760, and
760,760. Pixel 759,759 in the data now has RGB data from the same point in
the image plane.

Outing Trolls is FUN!

unread,
Nov 24, 2009, 4:01:56 PM11/24/09
to
On Tue, 24 Nov 2009 14:41:48 -0500, Alan Browne
<alan....@FreelunchVideotron.ca> wrote:

Listen, you amazingly and pathetically stooopid moron of a troll .... It's
not a matter of moving the colors over different photosites, but moving the
individual image details over each color of photosite. So that each
individual speck of image detail is then recorded by all 4 photosites in
each 2x2 RGGB group.

Can you start to grasp that in that amazingly ignorant and inexperienced
troll's mind of yours?

I thought not. But there might be others *almost* as stupid as you who are
believing your pathetically stupid bullshit and incomprehensible reasoning.

Alan Browne

unread,
Nov 24, 2009, 4:44:45 PM11/24/09
to

I understand that perfectly well. I don't understand why John A. has
both the sensor AND the CFA moving.

This is why only one of the sensor or array need move, not both.

However, the disadvantage of moving the sensor, rather than the CFA, is
that one has to correct for registration whereas if the CFA were moved,
then registration corrections are not needed. (Not that it's a big deal
considering the very slow shoot rate of the camera - and it's likely
this is done in raw processing on the computer and not in camera... hmm,
unless there's also a camera monitor view...

Statements regarding IS are a bit at the coarse level - IS stab in
camera moves the entire sensor chip carrier which includes the CFA.

What Hassy are doing is occurring within the chip carrier as far as I
can tell.

David J. Littleboy

unread,
Nov 24, 2009, 6:05:48 PM11/24/09
to

"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:eJudnTUfGKBTzpHW...@giganews.com...

> David J. Littleboy wrote:
>> "Alan Browne" <alan....@FreelunchVideotron.ca> wrote:
>>> John A. wrote:
>>>> I've seen it. They're moving the whole array. Otherwise each spot in
>>>> the image would still be captured with a Bayer array pattern that does
>>>> not achieve the desired effect.
>>>>
>>>> The filters *have to* move for this scheme to work. Moving the sensor
>>>> but not the filters does nothing to achieve the desired effect of
>>>> R+G+B capture at every pixel. They *are* moving the sensor so that
>>>> *must* include the filters.
>>> You don't seem to get that _relative_ to one another, one is moving,
>>> otherwise it's damned hard to get a different color over a pixel
>>> location.
>>
>> It's easy: they just shift the data. Assume pixel 759,759 in the sensor
>> is R. Shift the whole sensor shebang one pixel position left, and now
>> pixel 759,760 in the sensor (B) is measuring the same point in the image.
>> Shift vertically, and pixel 760,760 in the sensor (G) is measuring the
>> same point in the image.
>>
>> Now construct pixel 759,759 in the data from pixels 759,759, 759,760, and
>> 760,760. Pixel 759,759 in the data now has RGB data from the same point
>> in the image plane.
>
> I understand that perfectly well.

If that were true...

> I don't understand why John A. has both the sensor AND the CFA moving.

You wouldn't have said that.

> This is why only one of the sensor or array need move, not both.

Your head is completely wedged. Both the sensor and the filter array move
together.

Alan Browne

unread,
Nov 24, 2009, 6:29:04 PM11/24/09
to

That makes absolutely no sense. If a pixel sensor is to record first R,
then G, then B and then G again in separate shots, they certainly should
not move together.

See: http://www.hasselblad.co.uk/media/997168/uk_h3dii_ms39_datasheet_v3.pdf
in particular p. 3 where it says, as plainly as can be: "High precision

piezo motors control movements of the sensor in one pixel increments. By
combining four shots, each offset by one pixel, the true colours, Red,
Green and Blue of each point are obtained"

So all that moves is the sensor. The CFA stays fixed. There is even a
nice diagram to help you.


Wilba

unread,
Nov 24, 2009, 6:38:40 PM11/24/09
to
Alan Browne wrote:
>
> See:
> http://www.hasselblad.co.uk/media/997168/uk_h3dii_ms39_datasheet_v3.pdf
> in particular p. 3 where it says, as plainly as can be: "High precision
> piezo motors control movements of the sensor in one pixel increments. By
> combining four shots, each offset by one pixel, the true colours, Red,
> Green and Blue of each point are obtained"
>
> So all that moves is the sensor. The CFA stays fixed. There is even a
> nice diagram to help you.

Follow the X, which represents an RGB pixel at position 1, 1 in the final
image. In the first shot it's on sensor pixel 1, 1 (which is always red),
then it's on 1, 2 (always green), then 2, 2 (always blue), then 2, 1 (also
always green). Those four values are put together to get the RGB values for
the image pixel at position 1, 1.

If the sensor was moving under the filter array, the square with the X on it
would always be the same colour.


David J. Littleboy

unread,
Nov 24, 2009, 7:24:38 PM11/24/09
to

"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:P4GdnaSh7sbc8ZHW...@giganews.com...

Please read what I wrote above. It describes exactly how to do it.

> If a pixel sensor is to record first R, then G, then B and then G again
> in separate shots, they certainly should not move together.

They don't use the same physical pixel on the sensor to create the output
pixel, they use three different pixels in the sensor, all placed at the same
point in the sensor plane.

The idea you seem to be missing is the concept of constructing a single
output pixel from measurements taken with three different pixels.

> See:
> http://www.hasselblad.co.uk/media/997168/uk_h3dii_ms39_datasheet_v3.pdf
> in particular p. 3 where it says, as plainly as can be: "High precision
> piezo motors control movements of the sensor in one pixel increments. By
> combining four shots, each offset by one pixel, the true colours, Red,
> Green and Blue of each point are obtained"
>
> So all that moves is the sensor. The CFA stays fixed. There is even a
> nice diagram to help you.

The funniest thing here is that your idea that the CFA stays fixed is
completely dizzy. If the CFA doesn't move, each physical pixel position will
only see one color, and you wouldn't get a three-color image. This one is a
ROFL class mistake.

David J. Littleboy

unread,
Nov 24, 2009, 7:28:01 PM11/24/09
to

"Wilba" <use...@CUTTHISimago.com.au> wrote:
>
> Follow the X, which represents an RGB pixel at position 1, 1 in the final
> image. In the first shot it's on sensor pixel 1, 1 (which is always red),
> then it's on 1, 2 (always green), then 2, 2 (always blue), then 2, 1 (also
> always green). Those four values are put together to get the RGB values
> for the image pixel at position 1, 1.
>
> If the sensor was moving under the filter array, the square with the X on
> it would always be the same colour.

Exactly!

Alan Browne

unread,
Nov 24, 2009, 7:38:03 PM11/24/09
to
Wilba wrote:
> Alan Browne wrote:
>> See:
>> http://www.hasselblad.co.uk/media/997168/uk_h3dii_ms39_datasheet_v3.pdf
>> in particular p. 3 where it says, as plainly as can be: "High precision
>> piezo motors control movements of the sensor in one pixel increments. By
>> combining four shots, each offset by one pixel, the true colours, Red,
>> Green and Blue of each point are obtained"
>>
>> So all that moves is the sensor. The CFA stays fixed. There is even a
>> nice diagram to help you.
>
> Follow the X, which represents an RGB pixel at position 1, 1 in the final
> image. In the first shot it's on sensor pixel 1, 1 (which is always red),
> then it's on 1, 2 (always green), then 2, 2 (always blue), then 2, 1 (also
> always green). Those four values are put together to get the RGB values for
> the image pixel at position 1, 1.

I got the logic, just a different assumption about what was moving. See
below.

They're actually moving the sensor and array together and re-registering
the position separately. Less complex than what I was thinking but
still requires a lot of stability.

Apologies to John and David.

>
> If the sensor was moving under the filter array, the square with the X on it
> would always be the same colour.

Not if the CFA and the sensor array were _separate_ and moving wrt to
one another which was my (erroneous) assumption.

Alan Browne

unread,
Nov 24, 2009, 7:45:26 PM11/24/09
to

This means however, that other co's (SONY!!!!) can do the same thing if
they choose. The a900 would be a perfect candidate for a firmware
upgrade since it has the actuator in place.


Ray Fischer

unread,
Nov 24, 2009, 10:47:24 PM11/24/09
to

Go away, asshole troll.

--
Ray Fischer
rfis...@sonic.net

David J. Littleboy

unread,
Nov 24, 2009, 11:13:58 PM11/24/09
to

"Alan Browne" <alan....@FreelunchVideotron.ca> wrote:
>
> For still photography, every pixel is shot with each colour filter by
> mechanically moving the filter array by 1 pixel between shots. 4 shots
> later you have 50 Mpix with true RGB for each pixel. At that, it is 16
> bits/colour. No demosaicing means more accurate colour at each pixel and
> sharper images.
>
> For repros (of paintings), product, etc. this will be phenomenal.
>
> (At 1.4s per capture, an image will take nearly 6 seconds to take...
> better have a very sturdy tripod).

Not all that sturdy. As long as you lock up the mirror, it's amazing how
well an uninspired tripod does. I did a couple of interior projects this
year (no wind) with an ancient (almost 50 year old) Linholf aluminum tripod
with a tiny ball head (holding a 5D2 + Stigma 12-24), and exposures up to 30
seconds were fine.
...........

But now that we've got the technology figured out, here's what I find far
more interesting than the cutesy gimmick (which I predict won't improve
image quality noticeably).

http://www.hasselblad.co.uk/media/997168/uk_h3dii_ms39_datasheet_v3.pdf

"Moir� free images"

There are three possibilities here.
1. Hasselblad includes the low-pass filter that is mathematically required
for correct (and Moire free) imaging.
2. Hassleblad is lying.
3. Hasselblad is mathematically illiterate and doesn't know what Moire
means.

Take your pick...

(By the way, the lack of AA filters in MF backs has bitten a lot of people
really badly. These things are used in studio conditions with some of the
sharpest lenses ever made at optimal f stops with studio strobes, and the
moire makes fabrics impossible to photograph.)

David J Taylor

unread,
Nov 25, 2009, 2:30:47 AM11/25/09
to
"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:mdqdndizxqO745HW...@giganews.com...
[]

> This means however, that other co's (SONY!!!!) can do the same thing if
> they choose. The a900 would be a perfect candidate for a firmware
> upgrade since it has the actuator in place.

Exactly what I said many messages ago! But I also questioned what lenses
would have sufficient resolution to justify such a sub-sampling (of the
Bayer pattern).

Cheers,
David

David J. Littleboy

unread,
Nov 25, 2009, 3:30:51 AM11/25/09
to

"David J Taylor"
<david-...@blueyonder.not-this-bit.nor-this-part.co.uk.invalid> wrote:

> "Alan Browne" <alan....@FreelunchVideotron.ca> wrote:
> []
>> This means however, that other co's (SONY!!!!) can do the same thing if
>> they choose. The a900 would be a perfect candidate for a firmware
>> upgrade since it has the actuator in place.
>
> Exactly what I said many messages ago! But I also questioned what lenses
> would have sufficient resolution to justify such a sub-sampling (of the
> Bayer pattern).

It's not "sub sampling" anything. It's producing an image with exactly the
same resolution as the sensor.

David J Taylor

unread,
Nov 25, 2009, 3:57:43 AM11/25/09
to
"David J. Littleboy" <dav...@gol.com> wrote in message
news:ddudnZ7qjd2mdpHW...@giganews.com...
[]

> It's not "sub sampling" anything. It's producing an image with exactly
> the same resolution as the sensor.
>
> --
> David J. Littleboy
> Tokyo, Japan

One way of looking at it is that it is sub-sampling the Bayer group of
four pixels - in this case by making four exposures at half-Bayer pixel
increments.

The normal Bayer sensor does not produce an image at pixel resolution, but
only at a reduced resolution, even though what is read out /is/ at full
pixel resolution. The anti-alias filter in a well-designed camera will
have an idea point-spread function of somewhat more than a sensor pixel,
although admittedly somewhat less than full Bayer pixel resolution (i.e.
2x2 sensor pixels). In the Hasselblad arrangement the a less string AA
filter is required - nearer to a PSF of one sensor pixel.

Perhaps sub-sampling isn't the best word here, but I hope you see what I
mean.

Cheers,
David

David J. Littleboy

unread,
Nov 25, 2009, 5:22:25 AM11/25/09
to

"David J Taylor"
<david-...@blueyonder.not-this-bit.nor-this-part.co.uk.invalid> wrote in
message news:bk6Pm.8491$Ym4....@text.news.virginmedia.com...

> "David J. Littleboy" <dav...@gol.com> wrote in message
> news:ddudnZ7qjd2mdpHW...@giganews.com...
> []
>> It's not "sub sampling" anything. It's producing an image with exactly
>> the same resolution as the sensor.
>
> One way of looking at it is that it is sub-sampling the Bayer group of
> four pixels - in this case by making four exposures at half-Bayer pixel
> increments.

It's not at half-pixel increments. It takes four measurements at each pixel
location so that each pixel has all three colors (and green is oversampled).

"By capturing a sequence of 4 shots, each offset by a one pixel increment,
every point on your set is rendered with its true red, green and blue color
components."

> The normal Bayer sensor does not produce an image at pixel resolution,

Actually, it does. People just don't appreciate how incredibly cool Bayer
is.

> but only at a reduced resolution, even though what is read out /is/ at
> full pixel resolution. The anti-alias filter in a well-designed camera
> will have an idea point-spread function of somewhat more than a sensor
> pixel, although admittedly somewhat less than full Bayer pixel resolution
> (i.e. 2x2 sensor pixels). In the Hasselblad arrangement the a less string
> AA filter is required - nearer to a PSF of one sensor pixel.

In real life, Bayer AA filters are the same as what you'd use on a
monochrome camera. And that's because...

> Perhaps sub-sampling isn't the best word here, but I hope you see what I
> mean.

Bayer demosaicing reconstructs essentially the full resolution of the
sensor's pixel spacing for the luminance channel, and thus has the same
resolution requirements as a monochrome or full RGB per channel sensor. (If
Bayer were really noticeably worse than monochrome sensors in resolution,
digital photography would never have gotten off the ground. But the early
6MP cameras were as good or better than 35mm film the vast majority of the
time.)

The only time you see resolution advantages for Foveon is for the contrived
red on blue test charts. And it turns out that the human eye can't resolve
those either. (That is, Bayer is better than the human eye for color
resolution.)

David J Taylor

unread,
Nov 25, 2009, 6:07:42 AM11/25/09
to
"David J. Littleboy" <> wrote in message
news:C8GdnSWeQrb8mJDW...@giganews.com...
[]

> It's not at half-pixel increments. It takes four measurements at each
> pixel location so that each pixel has all three colors (and green is
> oversampled).

It's half a nominal Bayer pixel, a full sensor pixel.

>> The normal Bayer sensor does not produce an image at pixel resolution,
>
> Actually, it does. People just don't appreciate how incredibly cool
> Bayer is.

It produces an approximation, albeit a very good one.

[]


> In real life, Bayer AA filters are the same as what you'd use on a
> monochrome camera. And that's because...

[]


> David J. Littleboy
> Tokyo, Japan

In real life, there are a large number of variations in what AA filter is
used, varying right down to none!

At the moment, I find it an interesting development, but one in a very
specialised area. Whether it will be copied (is it patented?) by other
manufacturers for more normal uses is yet to be seen. I rather doubt that
it will have the appeal of, say, video-capable DSLRs.

Cheers,
David

David J. Littleboy

unread,
Nov 25, 2009, 6:20:28 AM11/25/09
to

"David J Taylor"

Since the math tells you that it provides essentially no improvement in
luminance resolution, it's appeal _should_ be extremely limited. (That's
"should" in the moral and engineering senses, as opposed to the predictive
sense.)

But most of the world is math-challenged, so there's lots of snakeoil that
can be sold.

--

David J Taylor

unread,
Nov 25, 2009, 6:53:34 AM11/25/09
to
"David J. Littleboy" <> wrote in message
news:GumdnUT_U4xlj5DW...@giganews.com...
[]

> Since the math tells you that it provides essentially no improvement in
> luminance resolution, it's appeal _should_ be extremely limited. (That's
> "should" in the moral and engineering senses, as opposed to the
> predictive sense.)
>
> But most of the world is math-challenged, so there's lots of snakeoil
> that can be sold.
>
> --
> David J. Littleboy
> Tokyo, Japan

Perhaps a little strong - if the images on their Web site are to be
believed. But not something I'll be saving up for!

Cheers,
David

They're Learning! Slow - But Learning Nonetheless!

unread,
Nov 25, 2009, 7:55:57 AM11/25/09
to
On Wed, 25 Nov 2009 20:20:28 +0900, "David J. Littleboy" <dav...@gol.com>
wrote:

They could be using sub-pixel shifts, but didn't want to have to explain
how that works in their press-releases. That would explain the extra
resolution as seen in their PDF brochure. People who buy expensive cameras
aren't necessarily intelligent (okay, let's be honest, they're rarely
intelligent). We've all seen proof of that. Explaining sub-pixel shifts and
resolution increases would have baffled and confused their buyers.

But at least they finally got this part right:

"The highly renowned HC/HCD lens line uses central lens shutters,
which adds flexibility by allowing flash to be employed at shutter
speeds up to 1/800s. The central shutter also improves image
quality by reducing camera vibration"

That's a vast improvement over *all* D/SLR camera designs.

It's one of the reasons I moved to super-zoom P&S cameras long ago. Finally
rid of that damnable and limiting focal-plane shutter. It's nice to see a
$32,000 digital camera finally learning from all the $100-$400 ones.
Advanced P&S camera owners have been enjoying noise-free, vibration-free,
distortion-free, flash-sync-limited-free cameras for a decade. At least one
camera line is finally catching up with the improvements afforded by doing
away with last century's archaic focal-plane shutter concept.

Now if only they could fix their flash-sync problem to allow for perfect
flash-sync up to shutter speeds as fast as 1/40,000 of a second, as all
CHDK capable P&S cameras can do for 3 years now. Maybe in their new and
improved $79,000 version next year. One can only hope.

Heavens forbid they should include all the other features of CHDK P&S
cameras too, like a user-programmable scripting system (in uBASIC and LUA),
user configurable cropping and alignment grids, et.al. (The list of CHDK
features that can't be implemented in any other camera designs is long.)
But then, advanced photographers' features like that would only confuse
their "intelligent" and wealthy buyers. LOL

BTW: Don't be too impressed over the number of megapixels. 50 megapixels is
only double the resolution of a 12.5 megapixel camera. Quite the markup for
just a 2-fold increase in image resolution. I can take 4 frames for
stitching them together in less time than this Hassy can do its pixel-shift
resolution trick automatically. Idiots are so easily swayed by a price-tag
and a brand-name.

Alan Browne

unread,
Nov 25, 2009, 1:44:02 PM11/25/09
to
David J. Littleboy wrote:
> "Alan Browne" <alan....@FreelunchVideotron.ca> wrote:
>> For still photography, every pixel is shot with each colour filter by
>> mechanically moving the filter array by 1 pixel between shots. 4 shots
>> later you have 50 Mpix with true RGB for each pixel. At that, it is 16
>> bits/colour. No demosaicing means more accurate colour at each pixel and
>> sharper images.
>>
>> For repros (of paintings), product, etc. this will be phenomenal.
>>
>> (At 1.4s per capture, an image will take nearly 6 seconds to take...
>> better have a very sturdy tripod).
>
> Not all that sturdy. As long as you lock up the mirror, it's amazing how
> well an uninspired tripod does. I did a couple of interior projects this
> year (no wind) with an ancient (almost 50 year old) Linholf aluminum tripod
> with a tiny ball head (holding a 5D2 + Stigma 12-24), and exposures up to 30
> seconds were fine.
> ...........
>
> But now that we've got the technology figured out, here's what I find far
> more interesting than the cutesy gimmick (which I predict won't improve
> image quality noticeably).
>
> http://www.hasselblad.co.uk/media/997168/uk_h3dii_ms39_datasheet_v3.pdf
>
> "Moiré free images"

>
> There are three possibilities here.
> 1. Hasselblad includes the low-pass filter that is mathematically required
> for correct (and Moire free) imaging.
> 2. Hassleblad is lying.
> 3. Hasselblad is mathematically illiterate and doesn't know what Moire
> means.

I took it to mean the moire-like effect that comes from de-mosaicing a
typical Bayer filter.

Alan Browne

unread,
Nov 25, 2009, 1:48:41 PM11/25/09
to

I have to apologize to you too then... sorry.

Forget zooms. Primes for the art work photography this system is aimed at.

I'd put my Minolta 100 f/2.8 Macro, Sony CZ 135 f/1.8, Hassy 120 f/4
Makro up to the task.

They are very sharp on the a900, and if the moire filter were put aside,
then the sharpness would really take off - at the risk of moire in some
images.

Alan Browne

unread,
Nov 25, 2009, 1:53:08 PM11/25/09
to
David J Taylor wrote:
> "David J. Littleboy" <dav...@gol.com> wrote in message
> news:ddudnZ7qjd2mdpHW...@giganews.com...
> []
>> It's not "sub sampling" anything. It's producing an image with exactly
>> the same resolution as the sensor.
>>
>> --
>> David J. Littleboy
>> Tokyo, Japan
>
> One way of looking at it is that it is sub-sampling the Bayer group of
> four pixels - in this case by making four exposures at half-Bayer pixel
> increments.

You have to watch terminology. You might say a "red" filter is a "half
- Bayer pixel" where to me and Littleboy a red filter is a whole pixel
whether or not it is filtered in color for subsequent de-mosaicing ...

David J Taylor

unread,
Nov 25, 2009, 2:17:48 PM11/25/09
to

"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:MY6dnc232oyU4ZDW...@giganews.com...
> David J Taylor wrote:
[]

>> Exactly what I said many messages ago! But I also questioned what
>> lenses would have sufficient resolution to justify such a sub-sampling
>> (of the Bayer pattern).
>
> I have to apologize to you too then... sorry.
>
> Forget zooms. Primes for the art work photography this system is aimed
> at.
>
> I'd put my Minolta 100 f/2.8 Macro, Sony CZ 135 f/1.8, Hassy 120 f/4
> Makro up to the task.
>
> They are very sharp on the a900, and if the moire filter were put aside,
> then the sharpness would really take off - at the risk of moire in some
> images.

Ah, lenses which cost about as much as my entire outfit! <G> But,
thanks, it's good to know that there are some lenses which might have
enough resolution. In theory, you still need an anti-alias filter, albeit
not as strong as with the Bayer sensor. Perhaps in practice the lenses
alone may be sufficiently low MTF....

Cheers,
David

David J Taylor

unread,
Nov 25, 2009, 2:21:11 PM11/25/09
to
"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:WuCdnf9zQPuI4JDW...@giganews.com...
[]

> You have to watch terminology. You might say a "red" filter is a
> "half - Bayer pixel" where to me and Littleboy a red filter is a whole
> pixel whether or not it is filtered in color for subsequent de-mosaicing
> ...

Yes, this does seem to have caused a little disagreement. The fundamental
unit of the Bayer mosaic is a four-pixel square array, but as David
pointed out the luminance resolution is greater - perhaps not one pixel,
but better than two pixels (in one direction). One problem for the new
system is that it's up against a Bayer sensor which has been well refined
over the years.

Cheers,
David

nospam

unread,
Nov 25, 2009, 3:30:22 PM11/25/09
to
In article <HsfPm.8675$Ym4....@text.news.virginmedia.com>, David J
Taylor
<david-...@blueyonder.not-this-bit.nor-this-part.co.uk.invalid>
wrote:

> > You have to watch terminology. You might say a "red" filter is a

> > "half - Bayer pixel" where to me and Littleboy a red filter is a whole
> > pixel whether or not it is filtered in color for subsequent de-mosaicing
> > ...
>
> Yes, this does seem to have caused a little disagreement. The fundamental
> unit of the Bayer mosaic is a four-pixel square array,

that's false. each pixel on a bayer sensor uses its neighbors to
calculate the missing components. they are not taken in clumps of four.

> but as David
> pointed out the luminance resolution is greater - perhaps not one pixel,
> but better than two pixels (in one direction). One problem for the new
> system is that it's up against a Bayer sensor which has been well refined
> over the years.

true

David J Taylor

unread,
Nov 25, 2009, 3:40:00 PM11/25/09
to

"nospam" <nos...@nospam.invalid> wrote in message
news:251120091530223102%nos...@nospam.invalid...

> In article <HsfPm.8675$Ym4....@text.news.virginmedia.com>, David J
> Taylor
[]

>> Yes, this does seem to have caused a little disagreement. The
>> fundamental
>> unit of the Bayer mosaic is a four-pixel square array,
>
> that's false. each pixel on a bayer sensor uses its neighbors to
> calculate the missing components. they are not taken in clumps of four.

The red or blue resolution cannot be better than the one red or blue pixel
per Bayer quad, so the colour resolution at least at the extremes of the
spectrum is that of the four pixel array - half that of the basic sensor
resolution. It gets more complex for green and luminance. It works well
because the eye has a lower spatial resolution for colour then for
luminance.

Cheers,
David

nospam

unread,
Nov 25, 2009, 4:32:20 PM11/25/09
to
In article <ACgPm.8712$Ym4...@text.news.virginmedia.com>, David J
Taylor
<david-...@blueyonder.not-this-bit.nor-this-part.co.uk.invalid>
wrote:

> >> Yes, this does seem to have caused a little disagreement. The

> >> fundamental
> >> unit of the Bayer mosaic is a four-pixel square array,
> >
> > that's false. each pixel on a bayer sensor uses its neighbors to
> > calculate the missing components. they are not taken in clumps of four.
>
> The red or blue resolution cannot be better than the one red or blue pixel
> per Bayer quad, so the colour resolution at least at the extremes of the
> spectrum is that of the four pixel array - half that of the basic sensor
> resolution. It gets more complex for green and luminance.

actually it can be better than one red or blue per quad. bayer
demosaicing looks to anywhere from 8 to 24 neighboring pixels (even
more but the benefits become very small). there are a number of ways to
do it, each with their own advantages and disadvantages.

however, with saturated red or blue, bayer will not do that well.
fortunately, that doesn't occur in the real world so it doesn't really
matter, except perhaps to foveon fans. if all you do is take photos of
red/blue charts, bayer is the wrong camera.

> It works well
> because the eye has a lower spatial resolution for colour then for
> luminance.

true. colour is sampled at half that of luminance but the eye can only
resolve 1/10th as much, so bayer is actually much better than what the
eye can resolve.

Wilba

unread,
Nov 25, 2009, 11:45:32 PM11/25/09
to

"If the sensor was moving under the filter array" means "the CFA and the
sensor array [are] _separate_ and moving wrt to one another".


David J Taylor

unread,
Nov 26, 2009, 3:49:25 AM11/26/09
to
"nospam" <nos...@nospam.invalid> wrote in message
news:251120091632207700%nos...@nospam.invalid...

> In article <ACgPm.8712$Ym4...@text.news.virginmedia.com>, David J
> Taylor
[]

>> It works well
>> because the eye has a lower spatial resolution for colour then for
>> luminance.
>
> true. colour is sampled at half that of luminance but the eye can only
> resolve 1/10th as much, so bayer is actually much better than what the
> eye can resolve.

Agreed - although whether it's 1/10th is probably subjective.

Cheers,
David

Kennedy McEwen

unread,
Nov 26, 2009, 5:22:28 AM11/26/09
to
In article <HuidnWNyZdeYMpHW...@giganews.com>, David J.
Littleboy <dav...@gol.com> writes

>
>
>But now that we've got the technology figured out, here's what I find far
>more interesting than the cutesy gimmick (which I predict won't improve
>image quality noticeably).
>
>http://www.hasselblad.co.uk/media/997168/uk_h3dii_ms39_datasheet_v3.pdf
>
>"Moir� free images"
>
>There are three possibilities here.
>1. Hasselblad includes the low-pass filter that is mathematically required
>for correct (and Moire free) imaging.
>2. Hassleblad is lying.
>3. Hasselblad is mathematically illiterate and doesn't know what Moire
>means.
>
>Take your pick...
>
>(By the way, the lack of AA filters in MF backs has bitten a lot of people
>really badly. These things are used in studio conditions with some of the
>sharpest lenses ever made at optimal f stops with studio strobes, and the
>moire makes fabrics impossible to photograph.)
>
I go with a variation of 1 & 3 ;-)

(3) They are, correctly, assuming that the main image damaging moir� is
the cross colour aliasing that derives from the spatial phase
differences between the RGB sampling in a Bayer filter. In that sense
it is "chroma moir� free", but not totally "moir� free".

(1) However, once all chroma moir� is eliminated even quite severe luma
moir� can be very acceptable and, in this case, it will actually be
quite low, since the lens MTF and the pixel area itself form a
reasonable luminance AA filter.

Now, if they were moving in fractions of a pixel in each axis then they
could eliminate both chroma and luma moir� - but that would require many
more frames in each shot - at least 8 and possibly 16. It depends how
effective the AA filter comprised in (1) above actually is, whether the
extra shots and increased image file size would be worth it.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

J. Clarke

unread,
Nov 26, 2009, 8:23:44 AM11/26/09
to

They mention "Moire removal technology" without saying what, specifically,
that technology might be.

You know, this is a capability that it should be easy to implement in any
camera with moving-sensor image stabilization--it should be a firmware
upgrade that on some could even be field-applied by the end user. And
there's no reason that the amount of shift can't be made
user-configurable--basically just have a menu option for the amount of shift
and the number of increments to shoot horizontally and vertically.

Alan Browne

unread,
Nov 26, 2009, 10:32:59 AM11/26/09
to

Well, Leica do without the AA filter in the M8 and M9 and there are not
too many complaints. Certainly with some subjects that will result in
moire. Although Leica claim to have an algorithm to deal with it.

Alan Browne

unread,
Nov 26, 2009, 10:39:49 AM11/26/09
to

Sorry, I still see the fundamental resolution as 1 pixel site or 1 CFA
site. De-mosaicing is imperfect of course resulting in colour
inaccuracies when pixel peeping.

OTOH, this camera is definitely aimed at large printing and pixel
peeping is almost a requirement.

Alan Browne

unread,
Nov 26, 2009, 10:43:11 AM11/26/09
to

I'm not sure that is really an issue when the point of this camera is
large repro prints. Red/Blue resolution will count.

David J Taylor

unread,
Nov 26, 2009, 10:45:06 AM11/26/09
to
"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:kL2dneUApt02ApPW...@giganews.com...
[]

> Well, Leica do without the AA filter in the M8 and M9 and there are not
> too many complaints. Certainly with some subjects that will result in
> moire. Although Leica claim to have an algorithm to deal with it.

Who would dare to complain when they had a Leica! <G>

Once sampled, you can't remove moir� with "an algorithm", although you may
be able to mask the effects under some circumstances. However, a fine
enough sampling and a low enough lens MTF at high spatial frequencies will
do.

Cheers,
David

David J Taylor

unread,
Nov 26, 2009, 10:47:51 AM11/26/09
to
"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:PJ6dnXK3o-PbPJPW...@giganews.com...
[]

> Sorry, I still see the fundamental resolution as 1 pixel site or 1 CFA
> site. De-mosaicing is imperfect of course resulting in colour
> inaccuracies when pixel peeping.

Yes, the spatial resolution is 1 sensor, but the colour resolution is not.
The luminance resolution is somewhere in between.

> OTOH, this camera is definitely aimed at large printing and pixel
> peeping is almost a requirement.

Indeed. I wonder what the first independent tests will show - I have a
feeling the brochure picture is simulated. What do you think?

Cheers,
David

Alan Browne

unread,
Nov 26, 2009, 10:50:56 AM11/26/09
to

Not in the scheme at hand. Which was my error.

Alan Browne

unread,
Nov 26, 2009, 10:56:07 AM11/26/09
to
J. Clarke wrote:
> Kennedy McEwen wrote:
>> In article <HuidnWNyZdeYMpHW...@giganews.com>, David J.
>> Littleboy <dav...@gol.com> writes
>>>
>>> But now that we've got the technology figured out, here's what I
>>> find far more interesting than the cutesy gimmick (which I predict
>>> won't improve image quality noticeably).
>>>
>>> http://www.hasselblad.co.uk/media/997168/uk_h3dii_ms39_datasheet_v3.pdf
>>>
>>> "Moiré free images"

>>>
>>> There are three possibilities here.
>>> 1. Hasselblad includes the low-pass filter that is mathematically
>>> required for correct (and Moire free) imaging.
>>> 2. Hassleblad is lying.
>>> 3. Hasselblad is mathematically illiterate and doesn't know what
>>> Moire means.
>>>
>>> Take your pick...
>>>
>>> (By the way, the lack of AA filters in MF backs has bitten a lot of
>>> people really badly. These things are used in studio conditions with
>>> some of the sharpest lenses ever made at optimal f stops with studio
>>> strobes, and the moire makes fabrics impossible to photograph.)
>>>
>> I go with a variation of 1 & 3 ;-)
>>
>> (3) They are, correctly, assuming that the main image damaging moiré

>> is the cross colour aliasing that derives from the spatial phase
>> differences between the RGB sampling in a Bayer filter. In that sense
>> it is "chroma moiré free", but not totally "moiré free".
>>
>> (1) However, once all chroma moiré is eliminated even quite severe
>> luma moiré can be very acceptable and, in this case, it will actually

>> be quite low, since the lens MTF and the pixel area itself form a
>> reasonable luminance AA filter.
>>
>> Now, if they were moving in fractions of a pixel in each axis then
>> they could eliminate both chroma and luma moiré - but that would

>> require many more frames in each shot - at least 8 and possibly 16.
>> It depends how effective the AA filter comprised in (1) above
>> actually is, whether the extra shots and increased image file size
>> would be worth it.
>
> They mention "Moire removal technology" without saying what, specifically,
> that technology might be.

It may be similar to what Leica claims for the M9 which does not have an
AA filter.


>
> You know, this is a capability that it should be easy to implement in any
> camera with moving-sensor image stabilization--it should be a firmware
> upgrade that on some could even be field-applied by the end user.


Indeed, I hope Sony go hat in hand to Hasselblad to license it.
However, it would be in Sony's greedy self interest to introduce it in a
new body rather than the existing cameras.

> And
> there's no reason that the amount of shift can't be made
> user-configurable--basically just have a menu option for the amount of shift
> and the number of increments to shoot horizontally and vertically.

I'd be very happy with the basic implementation.

Alan Browne

unread,
Nov 26, 2009, 11:03:58 AM11/26/09
to
David J Taylor wrote:
> "Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
> news:kL2dneUApt02ApPW...@giganews.com...
> []
>> Well, Leica do without the AA filter in the M8 and M9 and there are
>> not too many complaints. Certainly with some subjects that will
>> result in moire. Although Leica claim to have an algorithm to deal
>> with it.
>
> Who would dare to complain when they had a Leica! <G>
>
> Once sampled, you can't remove moiré with "an algorithm", although you
> may be able to mask the effects under some circumstances. However, a
> fine enough sampling and a low enough lens MTF at high spatial
> frequencies will do.

As discussed in other threads, I believe whatever their moire detection
is, that it is something beyond traditional signal sampling theory, more
akin to pattern detection and assumed correction, eg: some sort of "AI"
algorithm (to abuse both the notion and the term) that detects areas in
the image that have probably suffered moire and then guessing at the
original frequency and filling it in.

Alan Browne

unread,
Nov 26, 2009, 11:06:10 AM11/26/09
to

I think the brochure images are real. Hassy is a very conservative
company that hangs its hat on extremes of quality in commercial
photography. They would not stoop to simulation at the risk of not
being able to deliver in real life.

David J Taylor

unread,
Nov 26, 2009, 11:10:05 AM11/26/09
to
"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:Y5qdnanFxuNyO5PW...@giganews.com...
[]

> As discussed in other threads, I believe whatever their moire detection
> is, that it is something beyond traditional signal sampling theory, more
> akin to pattern detection and assumed correction, eg: some sort of "AI"
> algorithm (to abuse both the notion and the term) that detects areas in
> the image that have probably suffered moire and then guessing at the
> original frequency and filling it in.

Thanks, Alan. I don't understand why they don't just do the anti-aliasing
properly in the first place! But, there are those who like
over-sharpened, aliased images. Ho, hum!

Cheers,
David

David J Taylor

unread,
Nov 26, 2009, 11:11:47 AM11/26/09
to
"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:Y5qdnajFxuPvOpPW...@giganews.com...
> David J Taylor wrote:
[]

>> Indeed. I wonder what the first independent tests will show - I have a
>> feeling the brochure picture is simulated. What do you think?
>
> I think the brochure images are real. Hassy is a very conservative
> company that hangs its hat on extremes of quality in commercial
> photography. They would not stoop to simulation at the risk of not
> being able to deliver in real life.

I accept that viewpoint, but still await third-party tests. Admittedly
with an academic interest only, as I'm not in the market for that type of
camera.

Cheers,
David

Alan Browne

unread,
Nov 26, 2009, 2:12:10 PM11/26/09
to

I've seen a lot of non-AA'd images where there was no evidence of
aliasing - the subject was appropriate.

There is nothing wrong with "sharp!" and it can always be made softer.

David J Taylor

unread,
Nov 26, 2009, 2:56:31 PM11/26/09
to

"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:ZuKdndTB9OqXTpPW...@giganews.com...
[]

> I've seen a lot of non-AA'd images where there was no evidence of
> aliasing - the subject was appropriate.

Yes, if there's no high spatial frequency content to start with, there's
nothing to filter out.

> There is nothing wrong with "sharp!" and it can always be made softer.

I see so many over-sharpened images on the Internet I'm not sure I can
agree with you! The final image should have a clean dark-grey to
light-grey transition without any overshoot or ringing. That does mean it
may need to be processed differently for different display devices.
Over-sharpening can bring out excessive facial hair and other blemishes -
perhaps some people do that to be artsy, but I don't do artsy!

Cheers,
David

Alan Browne

unread,
Nov 26, 2009, 3:04:06 PM11/26/09
to

Over sharpening is ugly to be sure. One problem is that to optimally
sharpen one area you may be over or under sharpening another. Out comes
the selection brush and applying different degrees of sharpening to
different areas of the scene. Tedious. I generally USM for the area of
sharpness that is most important to the image. For a portrait that is
often the eyes, mouth, teeth, hair and in focus edges against the BG.
As these all have different line-frequency in their detail, the happy
compromise is usually to make the eyes the sharpest.

http://photo.net/photodb/photo?photo_id=8629589&size=lg

As to models (at least the ladies!), when they are fresh and young,
seeing their skin "real" with fine facial hair and pores looks wonderful
to me as long as it was exposed correctly - meaning more the lighting
setup than the camera settings alone ...

David J Taylor

unread,
Nov 27, 2009, 2:23:46 AM11/27/09
to
"Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
news:WO6dnUTQ3furQpPW...@giganews.com...
[]

> Over sharpening is ugly to be sure. One problem is that to optimally
> sharpen one area you may be over or under sharpening another. Out comes
> the selection brush and applying different degrees of sharpening to
> different areas of the scene. Tedious. I generally USM for the area of
> sharpness that is most important to the image. For a portrait that is
> often the eyes, mouth, teeth, hair and in focus edges against the BG. As
> these all have different line-frequency in their detail, the happy
> compromise is usually to make the eyes the sharpest.
>
> http://photo.net/photodb/photo?photo_id=8629589&size=lg
>
> As to models (at least the ladies!), when they are fresh and young,
> seeing their skin "real" with fine facial hair and pores looks wonderful
> to me as long as it was exposed correctly - meaning more the lighting
> setup than the camera settings alone ...

An excellent image, Alan, and one which suggests strongly that you know
what you are doing. Most of my stuff is not done in the studio, so I
don't often have the luxury of setting the lighting, but I do appreciate
that side lighting can bring out much better detail.

Cheers,
David

Alan Browne

unread,
Nov 27, 2009, 10:55:37 AM11/27/09
to
David J Taylor wrote:
> "Alan Browne" <alan....@FreelunchVideotron.ca> wrote in message
> news:WO6dnUTQ3furQpPW...@giganews.com...
> []
>> Over sharpening is ugly to be sure. One problem is that to optimally
>> sharpen one area you may be over or under sharpening another. Out
>> comes the selection brush and applying different degrees of sharpening
>> to different areas of the scene. Tedious. I generally USM for the
>> area of sharpness that is most important to the image. For a portrait
>> that is often the eyes, mouth, teeth, hair and in focus edges against
>> the BG. As these all have different line-frequency in their detail,
>> the happy compromise is usually to make the eyes the sharpest.
>>
>> http://photo.net/photodb/photo?photo_id=8629589&size=lg
>>
>> As to models (at least the ladies!), when they are fresh and young,
>> seeing their skin "real" with fine facial hair and pores looks
>> wonderful to me as long as it was exposed correctly - meaning more the
>> lighting setup than the camera settings alone ...
>
> An excellent image, Alan,

Thank you. But I can't take the credit. It was the camera, right? ;-)

> and one which suggests strongly that you know
> what you are doing. Most of my stuff is not done in the studio, so I
> don't often have the luxury of setting the lighting, but I do appreciate
> that side lighting can bring out much better detail.

The key advantage of a studio is setting the lighting range to the film
or sensor range and exposure consistency shot to shot. That out of the
way, you can concentrate on posing - which is very model dependent and
not easy at all.

Kennedy McEwen

unread,
Nov 27, 2009, 1:33:00 PM11/27/09
to
In article <helvt...@news5.newsguy.com>, J. Clarke
<jclarke...@cox.net> writes

>
>You know, this is a capability that it should be easy to implement in any
>camera with moving-sensor image stabilization--it should be a firmware
>upgrade that on some could even be field-applied by the end user. And
>there's no reason that the amount of shift can't be made
>user-configurable--basically just have a menu option for the amount of shift
>and the number of increments to shoot horizontally and vertically.
>
It doesn't need to be moving sensor IS, since the image is just being
moved relative to the CFA - which can be implemented with optical IS as
well.

Kennedy McEwen

unread,
Nov 27, 2009, 2:08:14 PM11/27/09
to
In article <zNydnWVsteSEOJPW...@giganews.com>, Alan Browne
<alan....@FreelunchVideotron.ca> writes

>
>Indeed, I hope Sony go hat in hand to Hasselblad to license it.

I don't think that Sony need to license this from anyone. The
Hasselblad implementation is certainly new, but the concept isn't.

Google "Swing CCD" and you will find lots of references to mainly
Japanese publications describing this technique. The earliest
publication I saw on it was over 25 years ago, by Sony themselves as a
means of reducing chroma aliasing in high resolution colour striped CCD
video cameras, and it wouldn't surprise me if it went back even further
than that.

It seemed to work quite well with video because the 4 field cycle of the
CCD movement matched the 8 field resync cycle of the chroma subcarrier
in both NTSC and PAL. Unlike the Hasselblad implementation no
interframe arithmetic to get the colour samples on each pixel was
required, just the viewer averaging out the equal and opposite chroma
phase errors on sequential fields in a similar way to PAL.

I never saw it actually implemented in anything other than lab
demonstration cameras though so there must have been some other issues.

Alan Browne

unread,
Nov 27, 2009, 7:53:46 PM11/27/09
to
Kennedy McEwen wrote:
> In article <helvt...@news5.newsguy.com>, J. Clarke
> <jclarke...@cox.net> writes
>>
>> You know, this is a capability that it should be easy to implement in any
>> camera with moving-sensor image stabilization--it should be a firmware
>> upgrade that on some could even be field-applied by the end user. And
>> there's no reason that the amount of shift can't be made
>> user-configurable--basically just have a menu option for the amount of
>> shift
>> and the number of increments to shoot horizontally and vertically.
>>
> It doesn't need to be moving sensor IS, since the image is just being
> moved relative to the CFA - which can be implemented with optical IS as
> well.

Good point. Not sure what the limitations might be, if any.

Message has been deleted

Bob Larter

unread,
Nov 28, 2009, 2:09:26 AM11/28/09
to
The P&S troll wrote:
> Now if only they could fix their flash-sync problem to allow for perfect
> flash-sync up to shutter speeds as fast as 1/40,000 of a second, as all
> CHDK capable P&S cameras can do for 3 years now.

HAHAhahahahahahahaha!

--
W
. | ,. w , "Some people are alive only because
\|/ \|/ it is illegal to kill them." Perna condita delenda est
---^----^---------------------------------------------------------------

Alan Browne

unread,
Nov 28, 2009, 10:47:58 AM11/28/09
to
Kennedy McEwen wrote:
> In article <zNydnWVsteSEOJPW...@giganews.com>, Alan Browne
> <alan....@FreelunchVideotron.ca> writes
>>
>> Indeed, I hope Sony go hat in hand to Hasselblad to license it.
>
> I don't think that Sony need to license this from anyone. The
> Hasselblad implementation is certainly new, but the concept isn't.
>
> Google "Swing CCD" and you will find lots of references to mainly
> Japanese publications describing this technique. The earliest
> publication I saw on it was over 25 years ago, by Sony themselves as a
> means of reducing chroma aliasing in high resolution colour striped CCD
> video cameras, and it wouldn't surprise me if it went back even further
> than that.
>
> It seemed to work quite well with video because the 4 field cycle of the
> CCD movement matched the 8 field resync cycle of the chroma subcarrier
> in both NTSC and PAL. Unlike the Hasselblad implementation no
> interframe arithmetic to get the colour samples on each pixel was
> required, just the viewer averaging out the equal and opposite chroma
> phase errors on sequential fields in a similar way to PAL.
>
> I never saw it actually implemented in anything other than lab
> demonstration cameras though so there must have been some other issues.

Interesting, in effect if there was a patent it has long since expired,
although a recent patent might have something particular in it that
refreshes the exclusivity.

Alan Browne

unread,
Nov 28, 2009, 10:53:32 AM11/28/09
to
John A. wrote:
> I think it's best done at the sensor. The adjustment is by definition
> fixed at the sensor, but if done in-lens would vary with both the
> length of the lens and the resolution of the sensor. Doing it in-lens
> just adds an element of complexity that does nothing to help the
> results. Doing it at the sensor, even coupled with in-lens IS you can
> just add very simple actuators to just move the sensor a single exact
> pixel width. In such an implementation, the IS system, in-lens or
> otherwise, doesn't even have to know about the multi-registration,
> making it compatible with any lens you've got.

I was thinking that optical path issues due to zoom and focus, and
perhaps manufacturing variances (both lens and body) would be the hard
part, beyond the angled rays.

Moving the sensor alone as a certain discrete simplicity about it.

It just occurred to me that some Hasselblad users might, with this
ability in the camera, be wishing that Hassy would go further and
implement IS for anti shake reasons. The range of motion is larger, to
be sure, but once the door is opened...

Kennedy McEwen

unread,
Nov 28, 2009, 12:41:45 PM11/28/09
to
In article <DpmdnU9W-qnh2ozW...@giganews.com>, Alan Browne
<alan....@FreelunchVideotron.ca> writes
>

>I was thinking that optical path issues due to zoom and focus, and
>perhaps manufacturing variances (both lens and body) would be the hard
>part, beyond the angled rays.
>
Those issues are already sorted out for the existing IS function, which
keeps the image stationary on the sensor irrespective of zoom, focus or
manufacturing tolerance. All that is needed is the camera to tell the
lens what the required displacement is, as the accelerometers currently
do.

>Moving the sensor alone as a certain discrete simplicity about it.
>

Sure, but there is a huge installed base of lenses with IS already built
in which could access this capability through a simple firmware or body
upgrade. Adding sensor motion mechanics to a camera that otherwise
doesn't need it adds cost.

Kennedy McEwen

unread,
Nov 28, 2009, 1:11:35 PM11/28/09
to
In article <86mdnY4NyoWz24zW...@giganews.com>, Alan Browne
<alan....@FreelunchVideotron.ca> writes
>Kennedy McEwen wrote:
>> In article <zNydnWVsteSEOJPW...@giganews.com>, Alan
>>Browne <alan....@FreelunchVideotron.ca> writes
>>>
>>> Indeed, I hope Sony go hat in hand to Hasselblad to license it.
>> I don't think that Sony need to license this from anyone. The
>>Hasselblad implementation is certainly new, but the concept isn't.
>> Google "Swing CCD" and you will find lots of references to mainly
>>Japanese publications describing this technique. The earliest
>>publication I saw on it was over 25 years ago, by Sony themselves
>
>Interesting, in effect if there was a patent it has long since expired,
>although a recent patent might have something particular in it that
>refreshes the exclusivity.

IANAIPL, but my understanding of patent law is that you can't refresh
exclusivity to a master patent just by filing new patents which
reference it with additional claims - it is only the new claims which
are protected from the later date. Indeed, I encountered this very
subject in exactly those circumstances in 1986 where the original Sony
publication was cited in one of my own patent applications.

Certainly the original Sony patents on this have long since expired, but
that doesn't stop them being cited as prior art since they would now be
considered public domain, thus preventing anyone else from obtaining
patent exclusivity of the basic technique. Similarly, patenting an
implementation of this by moving the image, instead of the sensor, would
find my own (also expired) and other patents being cited as prior art.
;-)

The Hasselblad implementation is new, but the concept isn't. I don't
think Sony need to license anything to implement this on their own
cameras - however, it is probably a much more valuable technique to
Hasselblad's target market than it is to Sony's.

Bob Larter is Lionel Lauer - Look it up.

unread,
Nov 28, 2009, 2:36:05 PM11/28/09
to
On Sat, 28 Nov 2009 17:09:26 +1000, Bob Larter <bobby...@gmail.com>
wrote:

>> Now if only they could fix their flash-sync problem to allow for perfect
>> flash-sync up to shutter speeds as fast as 1/40,000 of a second, as all
>> CHDK capable P&S cameras can do for 3 years now.
>
>HAHAhahahahahahahaha!

http://chdk.wikia.com/wiki/CameraFeatures


Bob Larter's legal name: Lionel Lauer
Home news-group, an actual group in the "troll-tracker" hierarchy:
alt.kook.lionel-lauer (established on, or before, 2004)
Registered Description: "the 'owner of several troll domains' needs a group where he'll stay on topic."

<http://groups.google.com/groups/search?hl=en&num=10&as_ugroup=alt.kook.lionel-lauer>

"Results 1 - 10 of about 2,170 for group:alt.kook.lionel-lauer."

Alan Browne

unread,
Nov 28, 2009, 3:51:53 PM11/28/09
to
Kennedy McEwen wrote:
> In article <DpmdnU9W-qnh2ozW...@giganews.com>, Alan Browne
> <alan....@FreelunchVideotron.ca> writes
>>
>> I was thinking that optical path issues due to zoom and focus, and
>> perhaps manufacturing variances (both lens and body) would be the hard
>> part, beyond the angled rays.
>>
> Those issues are already sorted out for the existing IS function, which
> keeps the image stationary on the sensor irrespective of zoom, focus or
> manufacturing tolerance. All that is needed is the camera to tell the
> lens what the required displacement is, as the accelerometers currently do.

I'm referring more to the ability to make a discrete one pixel offset
with reasonable integer accuracy. If it achieves 90% or better, it may
be enough. The point being that the IS/VR has no mechanical stopper
limit (at least not in the middle of its range) where I assume the hassy
sensor displacement does.

Further, the accelerometers are in the lens (IS/VR) and the camera
really has no "knowledge" (and doesn't need any) wrt to the IS/VR
system. This has implications for the installed base you refer to below.

>
>> Moving the sensor alone <h>as a certain discrete simplicity about it.


>>
> Sure, but there is a huge installed base of lenses with IS already built
> in which could access this capability through a simple firmware or body
> upgrade. Adding sensor motion mechanics to a camera that otherwise
> doesn't need it adds cost.

I wasn't suggesting that. Just that the Hassy solution, as is,
inherently has that benefit.

Alan Browne

unread,
Nov 28, 2009, 4:07:36 PM11/28/09
to
Kennedy McEwen wrote:
> In article <86mdnY4NyoWz24zW...@giganews.com>, Alan Browne
> <alan....@FreelunchVideotron.ca> writes
>> Kennedy McEwen wrote:
>>> In article <zNydnWVsteSEOJPW...@giganews.com>, Alan
>>> Browne <alan....@FreelunchVideotron.ca> writes
>>>>
>>>> Indeed, I hope Sony go hat in hand to Hasselblad to license it.
>>> I don't think that Sony need to license this from anyone. The
>>> Hasselblad implementation is certainly new, but the concept isn't.
>>> Google "Swing CCD" and you will find lots of references to mainly
>>> Japanese publications describing this technique. The earliest
>>> publication I saw on it was over 25 years ago, by Sony themselves
>>
>> Interesting, in effect if there was a patent it has long since
>> expired, although a recent patent might have something particular in
>> it that refreshes the exclusivity.
>
> IANAIPL, but my understanding of patent law is that you can't refresh
> exclusivity to a master patent just by filing new patents which
> reference it with additional claims - it is only the new claims which
> are protected from the later date.

Thanks for that clarification.

> Indeed, I encountered this very
> subject in exactly those circumstances in 1986 where the original Sony
> publication was cited in one of my own patent applications.
>
> Certainly the original Sony patents on this have long since expired, but
> that doesn't stop them being cited as prior art since they would now be
> considered public domain, thus preventing anyone else from obtaining
> patent exclusivity of the basic technique. Similarly, patenting an
> implementation of this by moving the image, instead of the sensor, would
> find my own (also expired) and other patents being cited as prior art. ;-)
>
> The Hasselblad implementation is new, but the concept isn't. I don't
> think Sony need to license anything to implement this on their own
> cameras - however, it is probably a much more valuable technique to
> Hasselblad's target market than it is to Sony's.

Good explanation and I do agree with the last bit ... but I would not
get upset if Sony offered a new body with the capability. The a900 is a
_very_ good studio camera and this added ability along with the better
lenses would show benefits in large prints.

Kennedy McEwen

unread,
Nov 29, 2009, 8:51:46 AM11/29/09
to
In article <8didnTKfnZH2EIzW...@giganews.com>, Alan Browne
<alan....@FreelunchVideotron.ca> writes
>Kennedy McEwen wrote:
>> In article <DpmdnU9W-qnh2ozW...@giganews.com>, Alan
>>Browne <alan....@FreelunchVideotron.ca> writes
>>>
>>> I was thinking that optical path issues due to zoom and focus, and
>>>perhaps manufacturing variances (both lens and body) would be the
>>>hard part, beyond the angled rays.
>>>
>> Those issues are already sorted out for the existing IS function,
>>which keeps the image stationary on the sensor irrespective of zoom,
>>focus or manufacturing tolerance. All that is needed is the camera
>>to tell the lens what the required displacement is, as the
>>accelerometers currently do.
>
>I'm referring more to the ability to make a discrete one pixel offset
>with reasonable integer accuracy. If it achieves 90% or better, it may
>be enough. The point being that the IS/VR has no mechanical stopper
>limit (at least not in the middle of its range) where I assume the
>hassy sensor displacement does.
>
It goes to the stop position when IS/VR is switched off, so there is a
stop position. Not that it needs a stopper - it just moves to one
position, then the next: it could be +/-0.5 pixels for all it matters.
Certainly on the Canon IS lenses, which I have had a close look at the
internals of, the position of the IS lens is constrained to an x-y
translation stage and the displacement in each axis controlled using a
coil in magnetic field - exactly the same as in-camera IS. I expect
that Nikon's VR is the same, but I haven't had a close look at the
internals of a Nikon lens.

Alan Browne

unread,
Nov 29, 2009, 11:35:47 AM11/29/09
to
Kennedy McEwen wrote:
> In article <8didnTKfnZH2EIzW...@giganews.com>, Alan Browne
> <alan....@FreelunchVideotron.ca> writes
>> Kennedy McEwen wrote:
>>> In article <DpmdnU9W-qnh2ozW...@giganews.com>, Alan
>>> Browne <alan....@FreelunchVideotron.ca> writes
>>>>
>>>> I was thinking that optical path issues due to zoom and focus, and
>>>> perhaps manufacturing variances (both lens and body) would be the
>>>> hard part, beyond the angled rays.
>>>>
>>> Those issues are already sorted out for the existing IS function,
>>> which keeps the image stationary on the sensor irrespective of zoom,
>>> focus or manufacturing tolerance. All that is needed is the camera
>>> to tell the lens what the required displacement is, as the
>>> accelerometers currently do.
>>
>> I'm referring more to the ability to make a discrete one pixel offset
>> with reasonable integer accuracy. If it achieves 90% or better, it
>> may be enough. The point being that the IS/VR has no mechanical
>> stopper limit (at least not in the middle of its range) where I assume
>> the hassy sensor displacement does.
>>
> It goes to the stop position when IS/VR is switched off, so there is a
> stop position.

Is that mechanically stopped or just a "no current in the coil" neutral
position?

> Not that it needs a stopper - it just moves to one
> position, then the next: it could be +/-0.5 pixels for all it matters.

With several qualifications, the most important being for the purpose of
registration, that wrt to the first image, the offset to the next image
must be close to a full pixel and so for all three additional images.
It's not the overall tolerance, but the shot to shot changes must be
close to 1 pixel in displacement.

> Certainly on the Canon IS lenses, which I have had a close look at the
> internals of, the position of the IS lens is constrained to an x-y
> translation stage and the displacement in each axis controlled using a
> coil in magnetic field - exactly the same as in-camera IS. I expect
> that Nikon's VR is the same, but I haven't had a close look at the
> internals of a Nikon lens.

Is there feedback for position in the coil control? If not, it would be
near impossible to get the 1 pixel shot to shot accuracy that is required.

The use of the piezo actuator in the hassy approach almost certainly is
mechanically constrained to 4 specific and nearly perfect 1 pixel grid
positions.

As to your assertion that this could be applied to all the IS/VR lenses
out there, there is first the communications issue (most IS/VR lenses
are stand alone from the camera in that respect) so the lenses would
have to be re-chipped to implement such, assuming there is a
communication path on the lens/body connection.

Kennedy McEwen

unread,
Nov 29, 2009, 12:49:58 PM11/29/09
to
In article <Po2dnZjIL_R5P4_W...@giganews.com>, Alan Browne
<alan....@FreelunchVideotron.ca> writes
>Kennedy McEwen wrote:
>
>> Not that it needs a stopper - it just moves to one position, then
>>the next: it could be +/-0.5 pixels for all it matters.
>
>With several qualifications, the most important being for the purpose
>of registration, that wrt to the first image, the offset to the next
>image must be close to a full pixel and so for all three additional
>images. It's not the overall tolerance, but the shot to shot changes
>must be close to 1 pixel in displacement.
>
What makes you think that controlling a stable current in the coil is
any more difficult for the optic than it is for the sensor? The IS can
control the coil current with enough precision to achieve much better
than a single pixel precision, which it must be capable of if IS is
going to work at all. Stability and precision are no different to
sensor-IS!

>> Certainly on the Canon IS lenses, which I have had a close look at
>>the internals of, the position of the IS lens is constrained to an
>>x-y translation stage and the displacement in each axis controlled
>>using a coil in magnetic field - exactly the same as in-camera IS. I
>>expect that Nikon's VR is the same, but I haven't had a close look at
>>the internals of a Nikon lens.
>
>Is there feedback for position in the coil control? If not, it would
>be near impossible to get the 1 pixel shot to shot accuracy that is
>required.
>

Of course there is feedback - do you really think IS works open-loop?

>The use of the piezo actuator in the hassy approach almost certainly is
>mechanically constrained to 4 specific and nearly perfect 1 pixel grid
>positions.
>

I disagree, there is no evidence at all that this is a mechanically
constrained system - mechanics wear out all too readily.

Also, you don't need anything close to "perfect" 1 pixel steps for this
to work well, even with very simple image reconstruction.

>As to your assertion that this could be applied to all the IS/VR lenses
>out there, there is first the communications issue (most IS/VR lenses
>are stand alone from the camera in that respect) so the lenses would
>have to be re-chipped to implement such, assuming there is a
>communication path on the lens/body connection.

There certainly is a comms path on Canon lenses, less certain of Nikon,
but I expect there is as well. All Canon lenses have programmable
firmware: pins 2,3, 4 form an I2C bus. Normally this is only used for
communication between the camera and lens for aperture and focus control
(camera to lens) and focal length (lens to camera) but it is also used
by Canon themselves to calibrate lenses to specific bodies.

Alan Browne

unread,
Nov 29, 2009, 1:29:13 PM11/29/09
to
Kennedy McEwen wrote:
> In article <Po2dnZjIL_R5P4_W...@giganews.com>, Alan Browne
> <alan....@FreelunchVideotron.ca> writes
>> Kennedy McEwen wrote:
>>
>>> Not that it needs a stopper - it just moves to one position, then
>>> the next: it could be +/-0.5 pixels for all it matters.
>>
>> With several qualifications, the most important being for the purpose
>> of registration, that wrt to the first image, the offset to the next
>> image must be close to a full pixel and so for all three additional
>> images. It's not the overall tolerance, but the shot to shot changes
>> must be close to 1 pixel in displacement.
>>
> What makes you think that controlling a stable current in the coil is
> any more difficult for the optic than it is for the sensor? The IS can
> control the coil current with enough precision to achieve much better
> than a single pixel precision, which it must be capable of if IS is
> going to work at all. Stability and precision are no different to
> sensor-IS!

I wasn't comparing to sensor-IS, but to the _discrete step_ piezo
electric approach that Hassy has. Further, my para above applies
equally to the sensor-IS approach as to the in lens IS approach. And by
the way, some in-camera IS _is_ based on piezo motors.

Absent a position measuring feedback mechanism, I don't believe that
controlling a coil that accurately (one position relative to the other)
is achievable.

>
>>> Certainly on the Canon IS lenses, which I have had a close look at
>>> the internals of, the position of the IS lens is constrained to an
>>> x-y translation stage and the displacement in each axis controlled
>>> using a coil in magnetic field - exactly the same as in-camera IS.
>>> I expect that Nikon's VR is the same, but I haven't had a close look
>>> at the internals of a Nikon lens.
>>
>> Is there feedback for position in the coil control? If not, it would
>> be near impossible to get the 1 pixel shot to shot accuracy that is
>> required.
>>
> Of course there is feedback - do you really think IS works open-loop?

It's not the "feedback" as much as "position feedback" that is
important. Coil feedback control can be current measurement but that
does not mean the position relative to another position is measured,
determined or controlled.

>
>> The use of the piezo actuator in the hassy approach almost certainly
>> is mechanically constrained to 4 specific and nearly perfect 1 pixel
>> grid positions.
>>
> I disagree, there is no evidence at all that this is a mechanically
> constrained system - mechanics wear out all too readily.

There's no evidence that it's not. But in any case, piezo electric
devices can be configured as two states and the mechanical change is a
fixed proportion of the piezo material dimension. Control the dimension
correctly and it is repeatable forever to a specific position without
mechanical stops.

>
> Also, you don't need anything close to "perfect" 1 pixel steps for this
> to work well, even with very simple image reconstruction.

I disagree. The objective is to measure RGB components separately in
order to provide accurate colours. So measuring the correct amount of
R, G and B at each registration position requires each pixel to be
accurately positioned at the same point for successive measurements.

This pixel registration accuracy seems to be borne out by Hassy's choice
of a piezo actuator.


>
>> As to your assertion that this could be applied to all the IS/VR
>> lenses out there, there is first the communications issue (most IS/VR
>> lenses are stand alone from the camera in that respect) so the lenses
>> would have to be re-chipped to implement such, assuming there is a
>> communication path on the lens/body connection.
>
> There certainly is a comms path on Canon lenses, less certain of Nikon,
> but I expect there is as well. All Canon lenses have programmable
> firmware: pins 2,3, 4 form an I2C bus. Normally this is only used for
> communication between the camera and lens for aperture and focus control
> (camera to lens) and focal length (lens to camera) but it is also used
> by Canon themselves to calibrate lenses to specific bodies.

I don't disagree, however, they will necessarily need to be re-chipped
(or updated) for this purpose - that was my point. (I don't believe
that Canon cameras have the ability to change the f/w in the lens as
part of a f/w update?).

Doug McDonald

unread,
Nov 29, 2009, 4:25:09 PM11/29/09
to
Alan Browne wrote:

>>
>>> The use of the piezo actuator in the hassy approach almost certainly
>>> is mechanically constrained to 4 specific and nearly perfect 1 pixel
>>> grid positions.
>>>
>> I disagree, there is no evidence at all that this is a mechanically
>> constrained system - mechanics wear out all too readily.
>
> There's no evidence that it's not. But in any case, piezo electric
> devices can be configured as two states and the mechanical change is a
> fixed proportion of the piezo material dimension. Control the dimension
> correctly and it is repeatable forever to a specific position without
> mechanical stops.
>

I see no reason why piezos and magnetics are any different in regard to
"2-state" ness. Both are continuously variable mechanical devices.
Piezos, per se, do not have quantized position: it is continuously variable
as the controlling voltage is changed (except for the integral nature of
charge, since what really controls is the charge.)

Doug McDonald

Message has been deleted

Kennedy McEwen

unread,
Nov 30, 2009, 2:38:39 AM11/30/09
to
In article <YrWdnb9-iJrnII_W...@giganews.com>, Alan Browne
<alan....@FreelunchVideotron.ca> writes
>

>Absent a position measuring feedback mechanism, I don't believe that
>controlling a coil that accurately (one position relative to the other)
>is achievable.
>
So we conclude that "absent a position measuring feedback mechanism" you
don't believe that optical IS works, since that relies on precise
knowledge of the image position on the focal plane at any instant.

>> Of course there is feedback - do you really think IS works open-loop?
>
>It's not the "feedback" as much as "position feedback" that is
>important. Coil feedback control can be current measurement but that
>does not mean the position relative to another position is measured,
>determined or controlled.
>

Since IS works, and has been proven to work for more than a decade, this
argument is moot.


>
>> Also, you don't need anything close to "perfect" 1 pixel steps for
>>this to work well, even with very simple image reconstruction.
>
>I disagree. The objective is to measure RGB components separately in
>order to provide accurate colours. So measuring the correct amount of
>R, G and B at each registration position requires each pixel to be
>accurately positioned at the same point for successive measurements.
>

Only for very simple reconstruction. Don't forget that simple
2-dimensional interpolation of 2 out of the 3 colour channels works very
well for a single Bayer exposure. Adding a further 3 exposures provides
another 3 dimensions in which to undertake that interpolation. We aren't
talking about a vast amount of additional information to be obtained -
perfect 1-pixel motion only ensures that the maximum error in the 2
interpolated colours is precisely zero, but it doesn't have to be zero
to be significantly better than 2-D interpolation.

As someone who has designed and measured similar systems for recovery of
full luminance response I can assure you that perfect registration is
certainly not necessary.


>
>I don't disagree, however, they will necessarily need to be re-chipped
>(or updated) for this purpose - that was my point. (I don't believe
>that Canon cameras have the ability to change the f/w in the lens as
>part of a f/w update?).

Don't think so, just need to accept a single additional piece of data
from the camera - pixel pitch.

John A.

unread,
Nov 30, 2009, 9:40:03 AM11/30/09
to
On Mon, 30 Nov 2009 07:38:39 +0000, Kennedy McEwen
<r...@nospam.demon.co.uk> wrote:

>In article <YrWdnb9-iJrnII_W...@giganews.com>, Alan Browne
><alan....@FreelunchVideotron.ca> writes
>>
>>Absent a position measuring feedback mechanism, I don't believe that
>>controlling a coil that accurately (one position relative to the other)
>>is achievable.
>>
>So we conclude that "absent a position measuring feedback mechanism" you
>don't believe that optical IS works, since that relies on precise
>knowledge of the image position on the focal plane at any instant.

Actually, knowledge of the focal length and input from the
accelerometer are probably sufficient.

[...]

>>I don't disagree, however, they will necessarily need to be re-chipped
>>(or updated) for this purpose - that was my point. (I don't believe
>>that Canon cameras have the ability to change the f/w in the lens as
>>part of a f/w update?).
>
>Don't think so, just need to accept a single additional piece of data
>from the camera - pixel pitch.

And all you need to do to get perfect toast is plug a video camera
into your old toaster. :)

http://xkcd.com/644/

Alan Browne

unread,
Nov 30, 2009, 9:40:55 AM11/30/09
to

Given that the hassy approach is for discrete amount of movement, and
that repeatability is desired, it is simpler than a coil/magnet (which
would need resistance as well to work against).

There are types of piezos that will move a precise amount for a given
control voltage.

Doug McDonald

unread,
Nov 30, 2009, 9:48:09 AM11/30/09
to
Alan Browne wrote:

>>
>> I see no reason why piezos and magnetics are any different in regard to
>> "2-state" ness. Both are continuously variable mechanical devices.
>> Piezos, per se, do not have quantized position: it is continuously
>> variable
>> as the controlling voltage is changed (except for the integral nature of
>> charge, since what really controls is the charge.)
>
> Given that the hassy approach is for discrete amount of movement, and
> that repeatability is desired, it is simpler than a coil/magnet (which
> would need resistance as well to work against).
>
> There are types of piezos that will move a precise amount for a given
> control voltage.

I see what you mean. Piezos are their own spring.

The point is not continuous variability but rather being their
own spring.

It is also true that piezos need no power if they are not moving.
Magnets do.

Doug McDonald

Alan Browne

unread,
Nov 30, 2009, 9:51:57 AM11/30/09
to
Kennedy McEwen wrote:
> In article <YrWdnb9-iJrnII_W...@giganews.com>, Alan Browne
> <alan....@FreelunchVideotron.ca> writes
>>
>> Absent a position measuring feedback mechanism, I don't believe that
>> controlling a coil that accurately (one position relative to the
>> other) is achievable.
>>
> So we conclude that "absent a position measuring feedback mechanism" you
> don't believe that optical IS works, since that relies on precise
> knowledge of the image position on the focal plane at any instant.

So you can confirm that there is _position_ feedback? What method please?


>
>>> Of course there is feedback - do you really think IS works open-loop?
>>
>> It's not the "feedback" as much as "position feedback" that is
>> important. Coil feedback control can be current measurement but that
>> does not mean the position relative to another position is measured,
>> determined or controlled.
>>
> Since IS works, and has been proven to work for more than a decade, this
> argument is moot.

So, you can confirm that in lens IS has positional feedback?

>>
>>> Also, you don't need anything close to "perfect" 1 pixel steps for
>>> this to work well, even with very simple image reconstruction.
>>
>> I disagree. The objective is to measure RGB components separately in
>> order to provide accurate colours. So measuring the correct amount of
>> R, G and B at each registration position requires each pixel to be
>> accurately positioned at the same point for successive measurements.
>>
> Only for very simple reconstruction. Don't forget that simple
> 2-dimensional interpolation of 2 out of the 3 colour channels works very
> well for a single Bayer exposure. Adding a further 3 exposures provides
> another 3 dimensions in which to undertake that interpolation. We aren't
> talking about a vast amount of additional information to be obtained -
> perfect 1-pixel motion only ensures that the maximum error in the 2
> interpolated colours is precisely zero, but it doesn't have to be zero
> to be significantly better than 2-D interpolation.

The whole point of the hassy approach is to avoid interpolation at all.
That is probably line 1 of the specification.

>
> As someone who has designed and measured similar systems for recovery of
> full luminance response I can assure you that perfect registration is
> certainly not necessary.

Again, the avoidance of any interpolation appears to be the Hassy
principle goal.


>>
>> I don't disagree, however, they will necessarily need to be re-chipped
>> (or updated) for this purpose - that was my point. (I don't believe
>> that Canon cameras have the ability to change the f/w in the lens as
>> part of a f/w update?).
>
> Don't think so, just need to accept a single additional piece of data
> from the camera - pixel pitch.

But to do so, the IS/VR function in the lens, which is a standalone
function, must be modified to accept that data.

Alan Browne

unread,
Nov 30, 2009, 9:54:15 AM11/30/09
to
John A. wrote:
> On Mon, 30 Nov 2009 07:38:39 +0000, Kennedy McEwen
> <r...@nospam.demon.co.uk> wrote:
>
>> In article <YrWdnb9-iJrnII_W...@giganews.com>, Alan Browne
>> <alan....@FreelunchVideotron.ca> writes
>>> Absent a position measuring feedback mechanism, I don't believe that
>>> controlling a coil that accurately (one position relative to the other)
>>> is achievable.
>>>
>> So we conclude that "absent a position measuring feedback mechanism" you
>> don't believe that optical IS works, since that relies on precise
>> knowledge of the image position on the focal plane at any instant.
>
> Actually, knowledge of the focal length and input from the
> accelerometer are probably sufficient.

The issue is accurate registration: precise x/y positioning of the IS
lens (lens axis is z) to offset by one pixel at the sensor. That is in
addition to the focus/zoom issue.

Alan Browne

unread,
Nov 30, 2009, 10:00:11 AM11/30/09
to
Doug McDonald wrote:
> Alan Browne wrote:
>
>>>
>>> I see no reason why piezos and magnetics are any different in regard to
>>> "2-state" ness. Both are continuously variable mechanical devices.
>>> Piezos, per se, do not have quantized position: it is continuously
>>> variable
>>> as the controlling voltage is changed (except for the integral nature of
>>> charge, since what really controls is the charge.)
>>
>> Given that the hassy approach is for discrete amount of movement, and
>> that repeatability is desired, it is simpler than a coil/magnet (which
>> would need resistance as well to work against).
>>
>> There are types of piezos that will move a precise amount for a given
>> control voltage.
>
> I see what you mean. Piezos are their own spring.

Yes, in a sense.

>
> The point is not continuous variability but rather being their
> own spring.
>
> It is also true that piezos need no power if they are not moving.
> Magnets do.

Piezos don't really consume much power regardless of state (is my
understanding).

Kennedy McEwen

unread,
Dec 1, 2009, 2:06:25 PM12/1/09
to
In article <vsednVhwOvTlRI7W...@giganews.com>, Alan Browne
<alan....@FreelunchVideotron.ca> writes
>

>Given that the hassy approach is for discrete amount of movement, and
>that repeatability is desired, it is simpler than a coil/magnet (which
>would need resistance as well to work against).
>
>There are types of piezos that will move a precise amount for a given
>control voltage.

Not over any reasonable range of temperatures there isn't, piezos are
notoriously temperature sensitive, since every piezo-electric material
is also pyro-electric.

However, a coil will move the same amount in a magnetic field
irrespective of temperature.

Kennedy McEwen

unread,
Dec 1, 2009, 2:48:18 PM12/1/09
to
In article <MtCdnVrh6IeQQY7W...@giganews.com>, Alan Browne
<alan....@FreelunchVideotron.ca> writes
>Kennedy McEwen wrote:
>> In article <YrWdnb9-iJrnII_W...@giganews.com>, Alan
>>Browne <alan....@FreelunchVideotron.ca> writes
>>>
>>> Absent a position measuring feedback mechanism, I don't believe that
>>>controlling a coil that accurately (one position relative to the
>>>other) is achievable.
>>>
>> So we conclude that "absent a position measuring feedback mechanism"
>>you don't believe that optical IS works, since that relies on precise
>>knowledge of the image position on the focal plane at any instant.
>
>So you can confirm that there is _position_ feedback? What method please?
>
I suggest you re-read what I wrote.

Your claim that this is impossible without position feedback is bunkum,
since at least as accurate control, if not more so, of the image on the
focal plane is required for IS to function. If IS can be achieved
without any *position* feedback, merely accurate drive current control
and sensing, then accurate position of the image on the focal plane can
be achieved too.

>>>> Of course there is feedback - do you really think IS works
>>>>open-loop?
>>>
>>> It's not the "feedback" as much as "position feedback" that is
>>>important. Coil feedback control can be current measurement but that
>>>does not mean the position relative to another position is measured,
>>>determined or controlled.
>>>
>> Since IS works, and has been proven to work for more than a decade,
>>this argument is moot.
>
>So, you can confirm that in lens IS has positional feedback?
>

Once again, read what I wrote. You are making a completely
unsubstantiated claim that position feedback is required - it isn't!


>
>The whole point of the hassy approach is to avoid interpolation at all.
>That is probably line 1 of the specification.
>

No, the whole point of the Hasselblad approach is to capture additional
information in the image - no probabilities about it. There are clues
in their sample image which indicates more is going on than is being
revealed in the data sheet.

It is loading more messages.
0 new messages