Olympus-OM
[Top] [All Lists]

[OM] Re: Sigma SD10

Subject: [OM] Re: Sigma SD10
From: Tim Hughes <timhughes@xxxxxxxx>
Date: Tue, 22 Jun 2004 06:14:17 -0700 (PDT)
Unfortunately this thread on Foveon is like many related to digital cameras, 
filled with confusing arcane aspects of digital signal processing and human 
perception!  Joe made a number of generally to the point comments and here are 
a few related. 
 
Joe wrote >>: It's not a marketing scam at all, and Carver Mead (the inventor ) 
 <<
 Agreed it is not a scam, but it is a complex engineering/human perception 
trade-off. Primary inventor for the X3 layered sensor is actually Merrill, who 
is a Foveon employee who used to work for National Semiconductor, and has a lot 
of integrated circuit patents for memories etc. as well.
 
>>(He Mead) ..is well known ...   Especially in analog circuits.<<
Actually he is probably best known for his work on automated Digital  Chip 
layout software tools and techniques. (And a classic influential book in the 
area, he co-authored with Lynn Conway who worked at PARC  (see 
http://ai.eecs.umich.edu/people/conway/Awards/Electronics/ElectAchiev.html). 
But more recently, he has done a lot of work on analog computing "simulations" 
of biological systems like  neuron,retina,ear etc.  
:http://www.ini.unizh.ch/telluride2000/Tell00/carver.htm

>>Joe wrote: Actually, a pox on all their houses.  It's all Marketing 
>>Megapixels. <<
Could not agree more. Foveon has not been very good at explaining their 
technology and has been guilty of doing some less than transparent marketing 
comparisons. 

>>Can you suggest some URLs?  I'd be interested; I've always wondered how well 
>>the Foveon sensor worked in practice. << 
The Dp Review is a reasonable comparison and comes to the conclusion that the 
SD10 3x3Mp competes favourably with Cannon's D10, 6Mp Beyer chips. But Phil 
Askey the reviewer, also argues in favour of retaining aliasing artifacts!
 
The more technical papers by Foveon authors (a bit biased) are good for people 
who like MTF/SFT curves. The comparisons are carefully chosen to make the 
improvement in performance possible for Foveon (vs Cannon etc), look much 
better than human perception might indicate. For example, the worst resolution 
condition for the Beyer chips is red-blue and the improvement factor they 
measure, is for R-B where the SFT curves (similar to MTF) for these colors have 
dropped to near zero contrast. A less biased comparison might be the 
improvement in spatial resolution at 50% contrast say.
http://foveon.com/docs/Frequency.pdf 
(Also for matlab test code: http://www.i3a.org/downloads.html
ISO12223 slant edge test)
An interesting comparison aspect dealt with in the paper is the fact that 
luminance information can be particularly good for Foveon. For a comparison 
with TV for example, the chrominanance information is usually coded at a lower 
resolution than luminance, so edges look good but color is actually lower 
resolution to take into account human perception isues.  I have always thought 
that at least in theory, the luminance information of a raw file for a Beyer 
sensor could be better than for Foveon, when converted to B&W image. In reality 
the Beyer color interpolated data is used in the raw file and so when combined 
to form B&W the lower resolution R-B blurs the higher res green. If the real 
uninterpolated data were seperated there would still be "noise" associated with 
color-greyscale mapping  mismatches between adjacent pixels. 
       A bit more philosphical document by Foveon related to photography 
history: 
http://foveon.com/docs/Century.pdf
There is another similar power point style version of this on the web 
somewhere, with the color photo examples etc.
A carefully selected comparison by Foveon highlights the weird aliasing 
artifacts possible from Beyer sensors:
http://foveon.com/X3_comparison.html
Unfortunately there is not enough information to know how apples to apples 
there comparison is. My cynical guess is, it is a 3Mp Beyer to 3x3Mp Foveon.
 
The main difference from the sigma SD9 to SD10 sensor seemed to be the addition 
of lenses on chip to capture a bit more light. 
 
Foveon has a new "F19" chip for industrial video/still applications, which has 
a lower resolution but can be electronically shuttered at video rates. 
 
For the same total R+G+B pixels i.e. a 3.3*3 MP Foveon versus a 10MP Beyer, the 
Foveon resolves 15% higher spatial resolution (lines/mm etc) for Red and Blue 
and 18% lower resolution than Beyer, for Green.  The argument for Beyer is the 
human eye has better green visual acuity so this improves *aparent resolution* 
more than 18% **on average colors**. Against this, the aliasing and weird moire 
is a lot worse for Beyer and the luminance resolution may be worse for Beyer, 
which might negate some of the acuity gain in aparent green resolution. If a 
fully effective R-B anti-alias filter is used, then it will reduce Beyer 
resolution for green, to that of blue and red, making Beyer guranteed worse 
than Foveon for all conditions! (To prevent this you would like an anti-alias 
filter to be wavelength dependent for Beyer, but this is likely impractical).  
Also any part of a picture which contains little or no green in it, will have 
lower resolution for a Beyer sensor than for a Foveon s
 ensor.
 
Regards,
Tim Hughes




==============================================
List usage info:     http://www.zuikoholic.com
List nannies:        olympusadmin@xxxxxxxxxx
==============================================

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