Sony A100 to A700

[A complimentary Cc of this posting was sent to
Alfred Molon
], who wrote in article :
In other words, this kit lens can overload the sensor in many settings.

I'm not sure 70% of the sensor line count is sufficient. But I'm not
deep enough in the math to be able to discuss this further.
In any case, if the sensor were full colour, you'd want to have an ideal
lens with 100% MTF until the line count of the sensor (and an ideal AA
filter with a rectangle response until Nyquist).
With a Bayer sensor you do the colour interpolation, and the situation
is different.

Yes, this "70%" figure is mostly related to the sensor being of Bayer
type. (See the comparison of Sigma vs Canikon in one of the latest
reviews on dpreview: it turns out that there is a significant "portion
of truth" in Sigma's "Foveon pixel count".)

On the other hand, 100%MTF is in no way necessary. Remember that
MTF=50% means "with proper postprocessing no information is lost, the
price being about 1/2 step decrease in S/N ratio".

Besides most if not all reviews of the Sony Alpha 350 (14MP, approx.
3000 lines) point out that the Sony 18-70 lens is a limiting factor and
state that a better lens is desperately needed.

Did not see anything even close to this said about SAL1680Z. As I
said already, the Alpha kit lens is slightly lousier than Canikon's.

Hope this helps,
Ilya

Ilya Zakharevich wrote:
[A complimentary Cc of this posting was sent to
Alfred Molon
], who wrote in article :
In other words, this kit lens can overload the sensor in many settings.
I'm not sure 70% of the sensor line count is sufficient. But I'm not
deep enough in the math to be able to discuss this further.
In any case, if the sensor were full colour, you'd want to have an ideal
lens with 100% MTF until the line count of the sensor (and an ideal AA
filter with a rectangle response until Nyquist).
With a Bayer sensor you do the colour interpolation, and the situation
is different.

Yes, this "70%" figure is mostly related to the sensor being of Bayer
type. (See the comparison of Sigma vs Canikon in one of the latest
reviews on dpreview: it turns out that there is a significant "portion
of truth" in Sigma's "Foveon pixel count".)

Not as significant as the fact that there's something fundamentally
wrong with the science behind the idea that you should devise a special
way of testing to show an "advantage" not seen using methods accepted
for many years.
Foveon "advantage" in resolving primary colour patterns is mainly
irrelevant to colour photography. Yes I know that foveon zealots take
delight in occasionally finding such patterns in flower petals etc in
order to try and prove a point. On balance, present foveon sensors
suck. The DP1 might get away with being okay as a novelty item, but the
dslrs don't.


On the other hand, 100%MTF is in no way necessary. Remember that
MTF=50% means "with proper postprocessing no information is lost, the
price being about 1/2 step decrease in S/N ratio".

Besides most if not all reviews of the Sony Alpha 350 (14MP, approx.
3000 lines) point out that the Sony 18-70 lens is a limiting factor and
state that a better lens is desperately needed.

Did not see anything even close to this said about SAL1680Z. As I
said already, the Alpha kit lens is slightly lousier than Canikon's.

Hope this helps,
Ilya

In article , Ilya Zakharevich
wrote:

I'm not sure 70% of the sensor line count is sufficient. But I'm not
deep enough in the math to be able to discuss this further.
In any case, if the sensor were full colour, you'd want to have an ideal
lens with 100% MTF until the line count of the sensor (and an ideal AA
filter with a rectangle response until Nyquist).
With a Bayer sensor you do the colour interpolation, and the situation
is different.

Yes, this "70%" figure is mostly related to the sensor being of Bayer
type. (See the comparison of Sigma vs Canikon in one of the latest
reviews on dpreview: it turns out that there is a significant "portion
of truth" in Sigma's "Foveon pixel count".)

there is no truth in the 'foveon pixel count.' pixels are pixels,
regardless what sigma/foveon claim (and they even use the term
inconsistently, further proof that their method is bogus).

the reason for the difference is that sigma omits the anti-alias filter
and thereby the sensor can resolve closer to nyquist, along with a lot
of aliasing and heavy sharpening, which some people mistake for real
resolution. it also does better on colour resolution charts but that
isn't anything that matters to human vision. it just makes bayer look
worse in an edge case that never occurs in nature.

[A complimentary Cc of this posting was sent to
frederick
], who wrote in article 1212458499.612758@ftpsrv1:
Yes, this "70%" figure is mostly related to the sensor being of Bayer
type. (See the comparison of Sigma vs Canikon in one of the latest
reviews on dpreview: it turns out that there is a significant "portion
of truth" in Sigma's "Foveon pixel count".)

Not as significant as the fact that there's something fundamentally
wrong with the science behind the idea that you should devise a special
way of testing to show an "advantage" not seen using methods accepted
for many years.

I suspect you may have very little clue in this topic. Read the
dpreview analysis (turns out that THIS topic is well within THEIR
level of technical expertise; I did not find any obvious goofs in what
they write). The conclusion is quite surprising...

[Keep in mind that they did not do any analysis of precision of
color reproduction; HERE Foveon should be, according to X3
documentation, quite bad.]

Hope this helps,
Ilya

[A complimentary Cc of this posting was sent to
nospam
], who wrote in article :
there is no truth in the 'foveon pixel count.' pixels are pixels,
regardless what sigma/foveon claim (and they even use the term
inconsistently, further proof that their method is bogus).

the reason for the difference is that sigma omits the anti-alias filter
and thereby the sensor can resolve closer to nyquist, along with a lot
of aliasing and heavy sharpening, which some people mistake for real
resolution. it also does better on colour resolution charts but that
isn't anything that matters to human vision. it just makes bayer look
worse in an edge case that never occurs in nature.

You can count angels at the point of a needle as long as you want.
The fact remains the fact.

DPreview (which is often suspected for prejudice in Canikon direction)
compares shots from a 5MP x 3 sensor, and from Bayer sensors with
(several) larger MP count. The conclusion was quite surprising (for me).

Hope this helps,
Ilya

Ilya Zakharevich wrote:
[A complimentary Cc of this posting was sent to
frederick
], who wrote in article 1212458499.612758@ftpsrv1:
Yes, this "70%" figure is mostly related to the sensor being of Bayer
type. (See the comparison of Sigma vs Canikon in one of the latest
reviews on dpreview: it turns out that there is a significant "portion
of truth" in Sigma's "Foveon pixel count".)

Not as significant as the fact that there's something fundamentally
wrong with the science behind the idea that you should devise a special
way of testing to show an "advantage" not seen using methods accepted
for many years.

I suspect you may have very little clue in this topic. Read the
dpreview analysis (turns out that THIS topic is well within THEIR
level of technical expertise; I did not find any obvious goofs in what
they write). The conclusion is quite surprising...

[Keep in mind that they did not do any analysis of precision of
color reproduction; HERE Foveon should be, according to X3
documentation, quite bad.]

Hope this helps,

It doesn't help. The defense for the DP review "analysis" is the words
"per pixel" resolution, and their noting of the fact that bayer sensors
of higher pixel count have (unsurprisingly except to Foveon zealots)
higher resolution.

Their conclusion also lists more "cons" than I've seen for any camera
for quite a while - as well as a low overall rating.

And I stand behind my comments above - it's dishonest / unscientific to
put the colour-resolution chart test in place of the normal chart tests.
If you wanted to show "optimum" performance in those charts, then
DPreview don't do it with other cameras (usually default jpeg output and
using old design lenses - not raw converted and optimised with OEM
converters and using the best possible lenses), so making an exception
for the unusual and generally not very good Sigma foveon cameras is only
helping to perpetuate myths.

Hope this helps.

[A complimentary Cc of this posting was sent to
frederick
], who wrote in article 1212478815.530753@ftpsrv1:
It doesn't help. The defense for the DP review "analysis" is the words
"per pixel" resolution, and their noting of the fact that bayer sensors
of higher pixel count have (unsurprisingly except to Foveon zealots)
higher resolution.

Wrong.

If you wanted to show "optimum" performance in those charts, then
DPreview don't do it with other cameras (usually default jpeg output and
using old design lenses - not raw converted and optimised with OEM
converters and using the best possible lenses), so making an exception
for the unusual and generally not very good Sigma foveon cameras is only
helping to perpetuate myths.

Now we know that it is not a myph. Visual comparison of Foveon images
and Bayer images gives an advantage of about 2x in pixel count.
(Myself, I was expecting something about 1.5x...)

Hope this helps,
Ilya

Ilya Zakharevich wrote:
[A complimentary Cc of this posting was sent to
frederick
], who wrote in article 1212478815.530753@ftpsrv1:
It doesn't help. The defense for the DP review "analysis" is the words
"per pixel" resolution, and their noting of the fact that bayer sensors
of higher pixel count have (unsurprisingly except to Foveon zealots)
higher resolution.

Wrong.

If you wanted to show "optimum" performance in those charts, then
DPreview don't do it with other cameras (usually default jpeg output and
using old design lenses - not raw converted and optimised with OEM
converters and using the best possible lenses), so making an exception
for the unusual and generally not very good Sigma foveon cameras is only
helping to perpetuate myths.

Now we know that it is not a myph. Visual comparison of Foveon images
and Bayer images gives an advantage of about 2x in pixel count.
(Myself, I was expecting something about 1.5x...)

Hope this helps,
Ilya
I don't know what you are arguing about.
4.69 x 2 = 10mp
IIRC DPreview already said the ("3.4mp x 3") SD10 was similar to 6mp
dslrs in resolution - so nothing new here.

"So what we'd really like to see is a sensor that combines the Foveon's
per pixel resolution with the output size of a modern DSLR. It would no
doubt take image quality onto another level and beat current Bayer
sensors by a mile."

Yes - sure - but not with noise etc.. By the time Foveon come up with
this (a "10mp x 3" sensor), do you expect that Bayer sensors will still
be where they are today? (or were yesterday - Sony, Canon, Nikon, Pentax
are already beyond 10mp).

In article , nospam says...

there is no truth in the 'foveon pixel count.' pixels are pixels,
regardless what sigma/foveon claim (and they even use the term
inconsistently, further proof that their method is bogus).

the reason for the difference is that sigma omits the anti-alias filter
and thereby the sensor can resolve closer to nyquist, along with a lot
of aliasing and heavy sharpening, which some people mistake for real
resolution. it also does better on colour resolution charts but that
isn't anything that matters to human vision. it just makes bayer look
worse in an edge case that never occurs in nature.

This has been discussed to death and there is no point to reopen the
discussion. Do a Google search on the archives to find out why with
Bayer sensors the effective resolution is lower than the nominal pixel
count.
--

Alfred Molon
------------------------------
Olympus 50X0, 8080, E3X0, E4X0, E5X0 and E3 forum at
http://tech.groups.yahoo.com/group/MyOlympus/
http://myolympus.org/ photo sharing site

In article 1212478815.530753@ftpsrv1, frederick says...

And I stand behind my comments above - it's dishonest / unscientific to
put the colour-resolution chart test in place of the normal chart tests.

Since the world is not black and white, it makes sense to use colour
resolution test targets instead of black and white ones.
--

Alfred Molon
------------------------------
Olympus 50X0, 8080, E3X0, E4X0, E5X0 and E3 forum at
http://tech.groups.yahoo.com/group/MyOlympus/
http://myolympus.org/ photo sharing site