Olympus-OM
[Top] [All Lists]

[OM] E1, 4/3 format and Foveon

Subject: [OM] E1, 4/3 format and Foveon
From: Hughes <hi100@xxxxxxxxxxxxx>
Date: Thu, 26 Jun 2003 06:14:53 -0700 (PDT)
Cc: chling@xxxxxxxxxxxxx, frugal@xxxxxxxxxxxxxxx
Andrew commented:
>>>The problem with this argument is that pixel count accounts for more than
one thing. If you are concerned with the colour fidelity, then the argument
holds water. Because each pixel captures full RGB, it is equivalent to a
sensor with 3x the number of pixels that only captures R, G, or B with each
pixel and then interpolates the results.

However, the other issue with pixel count is how big a print you can get
without it looking pixelated (or having to result to interpolation). The
larger pixel count image can be enlarged much larger than the smaller pixel
count image. It's also not a linear change either as the larger print would
be viewed from further away so a lower ppi print is still acceptable
(compared to a smaller print).
<<<

The color issue I have expanded on below. The issue of pixelation and 
interpolation are important
but remember the Bayer sensor software interpolates all its missing pixels 
before it stores the
image, while the Foveon does not have any missing pixels. A 1M Bayer sensor 
interpolates 0.5M
green pixels/sensors and 0.75M Red and 0.75M Blue pixels/sensors. So if the 
designers wanted to,
the Foveon camera could do a similar level of interpolation/upsampling and the 
result would be
just as smooth as higher pixel count Bayer.(But without the aliasing!) For 
memory consumption
reasons this probably still best done as postprocess or better still in the 
printer driver.

CH Ling commented:
>>Imaging resource suggest the SD9's resolution is less than
other 6MP DSLR:
http://www.imaging-resource.com/PRODS/SD9/SD9A12.HTM
<<
This is a different conclusion to the website referenced below but the issue is 
somewhat complex
and depends on what you test. (B&W testing helps improve Bayer sensors)

I did not really want to get sucked into the pixel wars but here is a quick 
example to illustrate
some of the issues:

For calculation convenience lets assume we have a 10mm*10mm sensor chip. We 
compare a 1M
(1000x1000 pixel)conventional Bayer pattern sensor to a 1M pixel (1000x1000 
sensor sites ) Foveon
X3 chip (3M R+G+B stacked sensors). We also assume there is no optical 
antialiasing filter used
and the lenses are ideal, not limiting resolution. 

If we start by using a **black and white** test target then the Nyquist limit 
for both cameras
could in theory be the same. That is neither camera can hope to resolve 
anything past the Nyquist
limit of 0.5*1000/10= 50 lines/mm. Any features with finer details would just 
produce low
frequency noise and moire. (The aliasing noise drops off as sine(X)/X ). In 
reality the Bayer chip
uses software to guess missing data. This means adjacent pixels typically get 
some interpolated
data reducing resolution below the B&W Nyquist limit and probably causing color 
fringing because
of the different pixel counts for each color. 

Now if we replace the black and white test chart with a green test chart then 
the resolution of
the Foveon chip is still 50 lines/mm, since the 3 color sensors are coincident. 
That is, the
Foveon chip has color independent resolution. 

For the Bayer sensor chip the resolution is lower for a green test chart, and 
even lower still,
for a red or a blue test chart. This is because (assume ideal color seperation) 
there are a lot
fewer green sensors than the Foveon chip.  The Nyquist resolution limit 
(ignoring interpolation)
would be: 
Green (707*707 sensors)         :  0.5*707/10 = 35.35 lines/mm  
Red or Blue (500x500 sensors ea):  0.5*500    = 25 lines/mm     
(with Bayer interpolation the numbers would be worse.)

To get the Green resolution to match that of Foveon we need at least a 2M Bayer 
sensor. 
To get the Red or Blue resolution to match that of Foveon we need at least a 4M 
Bayer sensor.  

See the DP review site for actual apples to apples comparison examples of the 
Foveon 3.4M pixel
(10.2M sensors) versus Cannon D30, 3.1M Bayer sensor:

http://www.dpreview.com/reviews/sigmasd9/page18.asp

Their summary :
" The EOS-D30 is still representative of the best of its sensor resolution, 
thus this is a good
comparison of X3 versus Bayer at the same output image size. As you can see the 
SD9 performs much
better than the EOS-D30, delivering more detail, sharper detail as well as 
elements which have
become invisible on the D30 image. That said the D30 does an admirable job 
considering the
considerable difference in the number of photo detectors, the SD9 has 10.3 
million, the D30 just
3.1 million. The important difference here is that where the D30's clever 
interpolations
algorithms have to guess details the X3 simply captures."

For a comparison of the 3.4M pixel Foveon to the 6M pixel Cannon D60 see the 
pictures at:
http://www.dpreview.com/reviews/sigmasd9/page19.asp

They state:
"As you can see the SD9 image looks sharp and detailed, it's difficult to find 
any part of the
EOS-D60 image which is actually exhibiting more detail than the SD9 and in some 
places the SD9
manages to pull out detail lost on the EOS-D60.
"

There is still further comparison with D60 on:
http://www.dpreview.com/reviews/sigmasd9/page20.asp
and comparisons to other cameras as well:
http://www.dpreview.com/reviews/sigmasd9/page23.asp
this comparison using a B/W test chart should favour D60 slightly but is 
actually quite favourable
to Foveon. Possibly because the D60 uses an antialias filter?
I think some of dpreviews conclusions are not strictly correct but... 
http://www.dpreview.com/reviews/sigmasd9/page24.asp.

A possibly less favourable issue for Foveon is that the color seperation 
"filters" are probably
not as sharp cutoff as other vendors. This leads to greater difficulty in color 
seperation and
could become a dynamic range issue between color channels. For example it might 
not be practical
to use seperate gain an offset for each color channel as is possible with Bayer 
sensors. (see
Nucore website for example of how this is done)

Regards,
Tim Hughes
TimHughes@xxxxxxxx















< This message was delivered via the Olympus Mailing List >
< For questions, mailto:owner-olympus@xxxxxxxxxxxxxxx >
< Web Page: http://Zuiko.sls.bc.ca/swright/olympuslist.html >


<Prev in Thread] Current Thread [Next in Thread>
Sponsored by Tako
Impressum | Datenschutz