Rawzor beta release (lossless raw compression)

Chris Malcolm wrote:
In rec.photo.digital.slr-systems Ray Fischer wrote:
Alan Browne wrote:
Dave Martindale wrote:
Alan Browne writes:

Please tell us that an AA filter does not blur an image.

Of course it blurs the image, in a particular way (it's not just a
Gaussian blur, or a disc blur). It does so in order to prevent
high-frequency information (fine detail) from being incorrectly resolved
as something it is not.

When a camera cannot resolve detail below a certain limit, do you think
it should

(a) substitute a blurred average, or
(b) substitute some other pattern that looks sharp but is not what was
present in the original image?

The AA filter is there to make the camera do (a) rather than (b). Many
people would argue that this is simply the correct thing to do, because
camera's shouldn't make up incorrect content - blurred is better than
wrong.

Other people would argue that (b) is sometimes better, because it looks
sharper, and no harm is done *unless you can tell that something is
wrong*. But that means you need to examine each image carefully,
because approach (b) can produce lots of subtle errors.

Thanks Dave. I'm sure Ray will find it very interesting.

Ray already knew all of that. It was part of the basics of those
computer graphics courses I took at Stanford. And anti-aliasing is
a lot harder in computer graphics.

I'd always supposed the AA problems in camera images and synthetic
images were the same. Why is it a much harder problem in computer
graphics?

Anti-aliasing a pixel means averaging all of the light that would be
in that pixel. In a camera you can simply blur together all the light
in a small region. In computer graphics you have to render the scene
that would be the pixel and that's a lot of additional time and work.
There are tricks and techniques, none of which I remember at the
moment, that speed up the process, but when the image is rendered you
can have infinite detail in a single pixel.

Imagine a picture of Manhattan. Now imagine if you could zoom in with
infinite detail into any window in the scene.

--
Ray Fischer

Ray Fischer wrote:
Wolfgang Weisselberg wrote:
Ray Fischer wrote:
Wolfgang Weisselberg wrote:
Ray Fischer wrote:

Is there some reason why microlenses don't do anti-aliasing?

Yes, they collect light into exactly one pixel, not spreading it
over more than one.

You're confusing "anti-aliasing" with "blurring".

Feel free to explain to me how an 2d optical low-pass filter
works without spreading the light over more than one pixel.

Feel free to show where I referred to a "2d optical low-pass filter".

Feel free to explain how anti-aliassing in an camera works without
being an 2d optical low-pass filter.

-Wolfgang

Chris Malcolm wrote:
In rec.photo.digital.slr-systems Wolfgang Weisselberg wrote:
Chris Malcolm wrote:

You can easily carry out your own experiments with AA filters by
downsizing a large sharp image which has patterned elements which will
produce aliasing artefacts at the downsize.

No, you cannot, unless you use an incompetent downsampling
algorithm. But you can test your downsampling method for
incomptence that way.

I wouldn't call it an incompetent downsizing algorithm, I'd call it a
simple one,

Ok, call it a simple one, if you like.

If I found that a downsizing method I was using was specifically
taking steps to remove aliasing artefacts without there being any way
I could switch that off I'd personally regard it as unsatisfactory.

If I found that a downsizing method produced aliasing artifacts
at /default/ settings, I'd dump it in a moment. I don't have
the inclination to re-invent the wheel and hand-hold an algorithm
for what should be --- and with competent implementations is ---
a trivial step.

If I want to roll my own algorithms, I can just as well implement
a stupid downsizing algorithm myself. It's not like nearest
neighbour is hard to implement ...

-Wolfgang

Wolfgang Weisselberg wrote:
Ray Fischer wrote:
Wolfgang Weisselberg wrote:
Ray Fischer wrote:
Wolfgang Weisselberg wrote:
Ray Fischer wrote:

Is there some reason why microlenses don't do anti-aliasing?

Yes, they collect light into exactly one pixel, not spreading it
over more than one.

You're confusing "anti-aliasing" with "blurring".

Feel free to explain to me how an 2d optical low-pass filter
works without spreading the light over more than one pixel.

Feel free to show where I referred to a "2d optical low-pass filter".

Feel free to explain how anti-aliassing in an camera works without
being an 2d optical low-pass filter.

How about a microlens array?

--
Ray Fischer

"Ray Fischer" wrote:

How about a microlens array?

The microlenses have a minimal AA effect, nowhere near enough to get things
under control. The second and later Sigma Foveon cameras have microlenses
and aliasing artifacts in spades.

--
David J. Littleboy
Tokyo, Japan

Ray Fischer wrote:
Paul Furman wrote:
Ray Fischer wrote:
Paul Furman wrote:
nospam wrote:

The point is that you can't remove the artefacts without removing
sensitivity to real non-artefactual detail as well.
Sure you can.

That's why AA
filters are always a compromise, with some makers setting them strong
enough to remove all aliasing artefacts, and others removing most, but
not all, so as to get a higher real detail resolution.
That's nonsense. An AA filter is not simply a blurring filter.

I wonder why they don't do that in software & leave the filter off.
Because it's impossible.

Here's the deal. You have a sensor on a chip surrounded by non-sensor
stuff. You want to get all of the light in the non-sensor area into
the sensor. That's the job of the AA filter.

You can't do it in software because light that doesn't make it into a
sensor is gone forever and nothing you can do in software will create
it.
So technically, you'd throw away data (light) by doing a gaussian blur
after the fact?

That wouldn't be an anti-aliasing operation.

Here's a visual example. The 'X' marks a sensor and a '.' is a
non-sensor area of the chip. Remeber that a tiny image falls upon
each one of the areas represented by the characters.

............
.X..X..X..X.
............

Any image falling on a '.' is lost. So what? If you have a pattern
such as (for example) closely spaced lines that line up with the 'X'
then you can get odd effects. If the matchup was exact then you'd get
all white because of the lines even if the lines were only 1/3rd of the
actual image. If you had three stripes of different colors then the
result would shift in color depending how the pattern lined up with
the sensors.

As you can see, blurring after the image has been captured has no
effect. There is nothing of the lost light to blur into the captured
image. Better would be to reduce the size of the non-sensitive area
and increasse the size of the sensitive area. That's not usually an
option because of circuit design limitations. So what they do is put a
filter of some kind in front of the sensors. Best would be a kind of
lens array that focuses all the light onto the sensors. Less good
would be something that would mix together all the light in an area so
that even if the sensor didn't get all of the light it would at least
get a uniform sample of the light.

Make sense? It's easier to describe with pictures.

Thanks, what I got from that is the AA filter captures the full image,
not just the part that squeaks past the microlenses because that version
is still missing some information but the AA filter gets it all.
Theoreticall perfect microlenses or 100% fill factor on the sensor (such
as backlit) would eliminate this advantage.

It still seems blurring after the fact should do pretty good though. But
I'd imagine that approach slows things down more.

--
Paul Furman
www.edgehill.net
www.baynatives.com

all google groups messages filtered due to spam

Ray Fischer wrote:
Wolfgang Weisselberg wrote:

Feel free to explain how anti-aliassing in an camera works without
being an 2d optical low-pass filter.

How about a microlens array?

And that wouldn't be a low pass filter why?
And that would work satisfactory how?

-Wolfgang

David J. Littleboy wrote:

"Ray Fischer" wrote:

How about a microlens array?

The microlenses have a minimal AA effect,

Idiot

--
Ray Fischer