The end of Bayer?

[David J. Littleboy]

...

Good example.

On the other hand a sensor without anti aliasing filter with a
random pattern would resove the sripes rather well - due to the
fact that the stripes are long and the brain will integrate over
the whole length of each stripe.

Actually, no. What happens without an antialiasing filter is that the system reports things that aren't there.

The test patterns that Phil took with the SD9 are an excellent example: although the pattern actually has 9 stripes, the camera shows 9 stripes, then 7 stripes, and then 5 stripes.

So with a random pattern, the sensor will report a random pattern that will be a different random pattern than the actuall pattern.

This is, of course, totally bogus behavior that has nothing to do with photography. I suppose it is artistic though: I remember a book on chess that collected artistic renditions of chess boards, and none of them had the right number of squares in the right positions.

This is also how film does it - and long thin lines are better
resolved with film than with digital sensors. NOTE - this has
nothing to do with that film resolves more than digital sensors -
it has only to do with tha fact that the sensor is random - and
that the lines are long.

Film is much more like a correctly antialiased digital sensor in that it doesn't report things that aren't there (except for grain patterns). It differs from most discrete sampling systems in the shape of its MTF curve (which has a long "tail" with extremely low response).

--
David J. Littleboy
Tokyo, Japan

 

[Roland Karlsson]

Actually, no. What happens without an antialiasing filter is that
the system reports things that aren't there.

Yes - I know that. But - how it reports it is important.

The test patterns that Phil took with the SD9 are an excellent
example: although the pattern actually has 9 stripes, the camera
shows 9 stripes, then 7 stripes, and then 5 stripes.

Not relevant though. Phil had no camera with a random sensor pattern.

So with a random pattern, the sensor will report a random pattern
that will be a different random pattern than the actuall pattern.

Yes - thats true. But - if the lines are long (like a powe line or such) then the human brain will se the line where it with a non random sensor will look just like som aliasing problem or even nothing at all.

Film is much more like a correctly antialiased digital sensor in
that it doesn't report things that aren't there (except for grain
patterns). It differs from most discrete sampling systems in the
shape of its MTF curve (which has a long "tail" with extremely low
response).

Not true. Film has exactly the same kind of aliasing problem as a random sensor. The grains has a position and I assure you that the grains were not there in the original picture. So - it definitely records things that were not there.

Roland

 

[Cesare Ferrari]

On the other hand a sensor without anti aliasing filter with a random pattern would resove the sripes rather well - due to the fact that the stripes are long and the brain will integrate over the whole length of each stripe.

I think this is an interesting idea, but without the application of an anti-alias filter you are still going to have problems. If we go back to considering the black/white striping idea the mathematically correct image for such a scene is a flat grey photo. Without an anti-aliasing filter your random pattern is bound to have photosites which 'see' just a white or just a black area of the scene, and hence you'll get some tonal variation across the scene. You'll probably end up with a blotchy image with which would appear to be sensor noise, but would in actual fact be the visual equivalent of jitter in an audio system.

This is also how film does it - and long thin lines are better resolved with film than with digital sensors. NOTE - this has nothing to do with that film resolves more than digital sensors - it has only to do with tha fact that the sensor is random - and that the lines are long.

Well, there's more to this than that. I'm not an expect in film, but i'd hazard a guess that there is a certain amount of scattering of light within the film, and this will lead to adjacent film grains getting some light from adjacent grains. In effect this is a 'built-in' anti-aliasing within the film...

Cesare

 

[Roland Karlsson]

Perhaps, but I've never seen that degree of "snap to grid" effect
with over-sharpened images from cameras with AA filters.

The pictures on those pages are made with a SD9. The SD9 has no micro lenses and the actual detector area is very small. Therefore you get a strong aliasing effect if you use sharp lenses.

The SD10 is a much better camera.

Roland

 

[Roland Karlsson]

Not if you design the CFA pattern differently.

Then it would not be a Bayer pattern.

Roland

 

[John Sheehy]

But - in the case of the Bayer
camera you get RGB directly instead of computed

Bayer is more direct, but typical conversion is not direct; RAW RGB colors from bayer sensors need varying hue shifts and contrast boosts to look like the real world color, but it is much closer to real saturation than the Sigmas are.

--
John

 

[John Sheehy]

Not me. If you have to measure all three colors at every point, you
need three times the data for essentially no improvement in image
quality.

You avoid color aliasing. Thats a big improvement for those picture
where it is evident. You also avoid the plastic filters that we
have no knowledge about how long they will keep their properties.

How long does a Foveon retain its properties?
--
John

 

[Roland Karlsson]

How long does a Foveon retain its properties?

Very long I assume - it is doped silicon - and you can find doped silicon in the nature that is hundreds of millions of years old.

The real threat to a sensor are corrosion to the metal parts, scratching and dirt I assume.

Roland

 

[Jack40001]

The term "A/D converter" below should be replaced by "DIGIC image processor". I will use an adaptive interpolation algorythm to reconstruct a full-color picture.

The A/D converter, which is aware of the layout of the Bayer filter, does > the blending of the various readings to determine the "color" for each pixel.

Jack.

 

[Kent D]

If it is a 3 ccd ssystem which is 4.4 MP per ccd, how many pixels does it have???
--
Kent Dooley

 

[Roland Karlsson]

If it is a 3 ccd ssystem which is 4.4 MP per ccd, how many pixels
does it have???

4.4 MP

 

[Jorge_Keys]

Full color pixel technology would be the only logical reason to upgrade at this time. Canon only has one or two lenses like the 135mm L that Bayer can "sort out" 11mp from. With the current crop of sensors in 95% of the cases greater than 8mp just renders a larger file with absolutly zero resolution benefits.

 

[Jorge_Keys]

Right now the 1D MKII is as good as it gets with a DSLR period. The only thing that would be a step up would be a MF back. Period.

The rest is just marketing yada, yada, yada.

 

[Z (is real)]

Very long I assume - it is doped silicon - and you can find doped
silicon in the nature that is hundreds of millions of years old.

Probably not a particularly good example as most of the long term reliability issues in integrated circuits involve phenomena that happen under circuit operation. If the chip is not powered up, it'll will likely last forever.

The real threat to a sensor are corrosion to the metal parts,
scratching and dirt I assume.

This is an incredibly well studied problem. Start by Googling "electromigration." (E.g. the front section of:
http://www.cadence.com/whitepapers/4095_Electromigration_WP.pdf
has a decent introduction.)

Don't have first hand info on the lifetimes targets for sensor designs, but my guess is camera electronics lifetime is very much dominated by discrete components, circuit boards, and chips other than the sensor. (The last because, compared to the sensor, the processors and memories use finer lithography, have more time in which at least some of their circuits are active, and operate with a higher power density, etc.)

The cynical part of me says that the electronics reliability of a Canon EOS-1 series camera is determined by the freakin' battery contacts

-Z-

 

[Roland Karlsson]

Right now the 1D MKII is as good as it gets with a DSLR period.
The only thing that would be a step up would be a MF back. Period.

The rest is just marketing yada, yada, yada.

Might be right.

There is one thing that might motivate more pixels - oversampling.

This is a common (always used actually) method for recording music. You sample at much higher frequency and then use some digital filtering. It could be used to improve on aliasing problems - at the cost of sensitivity.

Roland

 

[Roland Karlsson]

Don't have first hand info on the lifetimes targets for sensor
designs, but my guess is camera electronics lifetime is very much
dominated by discrete components, circuit boards, and chips other
than the sensor. (The last because, compared to the sensor, the
processors and memories use finer lithography, have more time in
which at least some of their circuits are active, and operate with
a higher power density, etc.)

Yepp.

This part of the thread discusses life time for a Foveon vs a Bayer.

I assume that the Foveon lasts just as long as any electronic chip.

I also assume that the colored filter in Bayer sensors last shorter.

I also assume that a digital sensor chip is not all that much powered up. So - the chip in the Foveon will last as long as it must last. Something else breaks before you see any color changes or the chip breaks.

Regarding the Bayer filter - I have no clue. 10 years - 100 years ... ???

The cynical part of me says that the electronics reliability of a
Canon EOS-1 series camera is determined by the freakin' battery
contacts

Or some of all the buttons.

Roland

 

[John Sheehy]

If it is a 3 ccd ssystem which is 4.4 MP per ccd, how many pixels
does it have???

4.4 MP, just like there is one John Sheehy if I stand in front of 3 mirrors hinged together; one red, one green, and one blue.

--
John

 

[John Sheehy]

Regarding the Bayer filter - I have no clue. 10 years - 100 years
... ???

If you take 20,000 images a year, at an average of 1/200s each at f/6.3 with lighting 1/4 the strength of sunlit objects, that's equal to 25 seconds a year of sunlit at f/6.3.

I don't think these sensors see much light!

--
John

 

[John Sheehy]

Full color pixel technology would be the only logical reason to
upgrade at this time. Canon only has one or two lenses like the
135mm L that Bayer can "sort out" 11mp from. With the current
crop of sensors in 95% of the cases greater than 8mp just renders a
larger file with absolutly zero resolution benefits.

You're way off in your assessment; there are lenses that can easily utilize 100MP full-frame. The 300 f/2.8L IS and 500mm f/4L IS, as two examples. I estimate 60-80 MP FF for my Tamron 90mm f/2.8 Di Macro.

People use the 500mm f/4L IS with 2x and 1.4x teleconverters stacked, with decent pixel-to-pixel contrast. That's like using a sensor with almost 9x as many pixels; actually better, because the TC's optical problems are not present.

--
John

 

[John Sheehy]

This is a common (always used actually) method for recording music.
You sample at much higher frequency and then use some digital
filtering. It could be used to improve on aliasing problems - at
the cost of sensitivity.

You also get better digital effects like pitch shifting when you oversample frequency.

--
John