Sensor ?

[1chaz]

If the sd 14 is considerd a 4.7 mp sensor (capturing 3 colors at each pixel ) then wouldn"t a 12 mp beyer sensor really be a 4.0 mp X 3 colors = 12 ???.. I have a new sd 14 with a 18x50 ex and also own a 10d and a 20d and have taken identical shots and can see why people love this sensor yes the body is not as perfected or polished as a canon or nikon but that iq is impressive and the color is spot on. Such large files for a 4.7 mp lots o bits

 

[Zone8]

Please take my advice and forget all the hype, nonsense etc. about MP and just enjoy using the camera to make high quality images. As one related example - how many people argued about or even wanted to know about how film emulsions were constructed.

That's what it really is about and this topic brings forth lots of venom, as already doing (time and again) at present on another forum.

--
Zone8

The photograph isolates and perpetuates a moment of time: an important and revealing moment, or an unimportant and meaningless one, depending upon the photographer's understanding of his subject and mastery of his process. -Edward Weston
http://www.photosnowdonia.co.uk/ZPS

 

[gbainbridge]

If the sd 14 is considerd a 4.7 mp sensor (capturing 3 colors at each
pixel ) then wouldn"t a 12 mp beyer sensor really be a 4.0 mp X 3
colors = 12 ???..

Not quite. The Beyer sensor has twice as many green pixels than red or blue so a 12 M(sub)pixel Beyer sensor can produce a true luminance resoution equivalent to 6 MPixels, although it only has a colour resolution of 3 Mpixels. That however ignores the anti-alias filter and to totally avoid any colour moiré effects the AA filter has to reduce the resolution to the equivalent of the lower value of 3 MPixels. In practice no manufacturers actually use such a severe AA filter, so it's always a compromise between resolution and moiré reduction.

As a result most monochrome resolution charts show the Foveon chip to have a resolution similar to that of a 10 Mpixel beyer chip, but use of a coloured resolution chart gives the game away.

See the comparison between the DP1 and vaiousl Beyer based cameras on the last page of the dpreview review for examples ( http://www.dpreview.com/reviews/sigmadp1/page20.asp ).

--
Thanks,
Gary.

 

[light_reaction]

Bayer 12MP:
6MP green, 3 MP Red, 3 MP Blue (input)
Foveon 4.7 MP:
4.7MP Green, 4.7 MP Red, 4.7 MP Blue (raw conversion output)

The bayer records more or less the actual luminance information per color and the Foveon records three color subspectri (not really rgb) which are then transformed during raw conversion.

What makes the difference is the lack of an AA filter on the foveon and the (unchangeable) fact that the resolution is the same for each channel and not different like with Bayer sensors. Both approaches have weaknesses and strengths so it is great for you to own a representative of each technology.

O.
--
http://www.flickr.com/photos/ollivr/
http://www.flickriver.com/photos/ollivr/popular-interesting/
http://seen.by.spiegel.de/ollivr-1

 

[Paul Petersen]

I tend to let the pictures do the talking, once you see the difference its hard to go back. But I still keep my Nikon D70 for the low light high ISO stuff the Sigma can't do.

Although it would be nice to have a Sigma camera system that is as well sorted out as my Nikon. I use the Nikon when I don't want to think about the camera.
Paul
--
A bad day of train chasing is better than a good day at work.
http://peterzpicts.smugmug.com/

 

[Ed_S]

This is scary - it really happened. I had a dream - no identifiable source of the "inspiration" - over the weekend. I had a number of props and was going to explain CFA vs Foveon in a "lecture" or "workshop" setting. Sufficiently intense that I was actually ruminating (in the dream) about choices of terms - how best to explain it, etc. Maybe Ole can interpret it for me. In any case - I think it's too much exposure to this issue.

From several years of debating this issue on this forum and elsewhere, there is the potential for at least as many points of view as you encounter those willing to chime in. Different folks have their own views and interpretations of the numbers - some see "the effect" others don't or discount it. The advice already given in this thread is good, to include "just enjoy it."

Kind regards,
--
Ed_S
http://www.pbase.com/ecsquires

 

[Roland Karlsson]

Foveon 4.7 MP:
4.7MP Green, 4.7 MP Red, 4.7 MP Blue (raw conversion output)

Personally I think it is more appropriate to view it as:
4.7 MP bluish (top layer)
4,7 MP luminosity (middle layer)
4.7 MP reddish (bottom layer)

Then you have to take the conversion algorithm into account. According to a white paper on Foveon´s web pages - the conversion is based on using the middle layer for luminosity and then extracting downsampled difference channels for color information. This means that the color resolution for Foveon based cameras (if they use the algorithm described by Foveon) does not have full color resolution, i.e. the color resolution is less than 4.7 MP.

--
Roland

 

[light_reaction]

Foveon 4.7 MP:
4.7MP Green, 4.7 MP Red, 4.7 MP Blue (raw conversion output)

Personally I think it is more appropriate to view it as:
4.7 MP bluish (top layer)
4,7 MP luminosity (middle layer)
4.7 MP reddish (bottom layer)

Then you have to take the conversion algorithm into account.

That's why I distinguished between "Input/bayer" and "output/foveon". Also, some time ago when we were discussing Pavel's and your tools for deriving raw output, it became clear that r/g/b can be derived by simply applying auto curves/levels and saturation adjustment.

According to a white paper on Foveon´s web pages - the conversion is
based on using the middle layer for luminosity and then extracting
downsampled difference channels for color information. This means
that the color resolution for Foveon based cameras (if they use the
algorithm described by Foveon) does not have full color resolution,
i.e. the color resolution is less than 4.7 MP.

That may well be true but if you apply such a rather strict operationalization then it becomes even more difficult to rate bayer sensors, where not only the pixel count is different but, from an isometric perspective, also the distance between pixels is different between g pixels and r/b pixels. This is what makes you get more apparent resolution in the output image for vertical and horizontal items and less apparent resolution for diagonal detail, while it is the same amount for both cases with foveon sensors.

To compare full color resolution, one'd need to take interval shots of a resolution chart with constantly changing color. Take a nice four-digit amount of frames and run them through an analysis software, then you can get real resolution figures (as long as you use the same lens type on each camera and none of the different mount copies is flawed in some way). Even here, this would leave out of account that different raw conversion routines are used which may also affect resolution.

Therefore, I think it is difficult enough to find a better suited measure by which to compare the two sensor types that it is reasonable practice to keep comparing the uninterpolated but r/g/b states - Bayer after capture and Foveon after conversion. By doing so, one smoothly gets down to the bottom line: neither the mainstream bayer, nor the boutique bayer (Kodak or Fuji if you will), nor Foveon is perfect. Each is worse than the other under certain circumstances.

O.
--
http://www.flickr.com/photos/ollivr/
http://www.flickriver.com/photos/ollivr/popular-interesting/
http://seen.by.spiegel.de/ollivr-1

 

[Ed_S]

Roland, O, and all,

I think this brings us back to the lack of a really common system of metrics. It's hard to "equate" pixel counts and what they represent between the two systems (i.e. CFA and Foveon). If I recall, somewhere back in our collective skull sessions, we posited the idea that each of the two systems almost merit their individual metrics, but that the current CFA based system is so entrenched as to lend great inertia to any change. I think we even suggested that marketeers and individual salespersons (have to blame someone) would resist trying to really explain (without bias) the bean counting for the two very different systems.

Ultimately, the Foveon (oversimplified) is Xmp x R + G + B and Bayer is Xmp = X/2G + X/4R + X/4B. (You're welcome to correct the algebra). Then factor in anti-aliasing or no, etc.

Kind regards,
--
Ed_S
http://www.pbase.com/ecsquires

 

[Sigmagus]

Hi! So you are the one who owns both SD14 & K20d, then I've been looking for you for a long time. Could you, please, tell us about IQ of K20d in comparison to SD14? Does it differ too much in more ample treatment i.e. at ISO100? Tell us more about your feelings. We can discuss sensors dimentions and calculate MP using algebra or even cybernetics, but finally we post here images not figures and percieve them with eyes and hearts. Most of the images are great and excite. All in all, can K20d IQ excite like SD14 does? Do you use SD14 as the first DSLR (for pleasure) and K20d as the second one (for trips) or quite the contrary?
Kind Regards,
Oleg Gusev
--
http://picasaweb.google.ru/Sigmagus

 

[Roland Karlsson]

Roland, O, and all,

I think this brings us back to the lack of a really common system of
metrics. It's hard to "equate" pixel counts and what they represent
between the two systems (i.e. CFA and Foveon). If I recall, somewhere
back in our collective skull sessions, we posited the idea that each
of the two systems almost merit their individual metrics, but that
the current CFA based system is so entrenched as to lend great
inertia to any change. I think we even suggested that marketeers and
individual salespersons (have to blame someone) would resist trying
to really explain (without bias) the bean counting for the two very
different systems.

Ultimately, the Foveon (oversimplified) is Xmp x R + G + B and Bayer
is Xmp = X/2G + X/4R + X/4B. (You're welcome to correct the algebra).
Then factor in anti-aliasing or no, etc.

Oh .... I was not talking about the MP count. It is quite uninteresting to compare MP count between the two solutions IMHO.

And I was not talking about Bayer CFA. It is VERY complex to analyze the full system of AA filter, Bayer CFA filtered sensor and reconstruction algorithm.

I was only trying to give some light on how the Foveon system works. It is not so simple as most makes it. It is very far from an RGB sensor. There are complex things going on in the conversion, both doing strange stuff with the color space and also smoothing the color resolution. But ... it IS possible to visualize what this process does .... contrary to the Bayer CFA reconstruction which is by far too complex.

--
Roland

 

[Ed_S]

Roland,

Thanks for the additional perspective.

Actually, I didn't really intend to come across as offering a critique or counterpoint of your or others' explanations up to that point. It was just musing on my part about this thread and past conversations as to how difficult it seems to be to create a consensus "lingua franca" for the two systems.

Kind regards,
--
Ed_S
http://www.pbase.com/ecsquires

 

[Roland Karlsson]

Ed_S wrote:
It was just musing on my part about this thread and past

conversations as to how difficult it seems to be to create a
consensus "lingua franca" for the two systems.

Yeah - it is hard.

Most people (me included) want some numbers for comparison; if possible just a few or even only one. This makes life simpler.

But sometimes it is not possible - either for pure technical reasons or because different people have different opinions how to compute the number(s). In the case of Foveon - both reasons apply.

My opinion i that the MP number is an interesting parameter for a camera. And my opinion is also that both Bayer CFA and Foveon is best measured per spatial photo site - mainly because that is the best way to map to an RGB image. Then the SD14/DP1 is 4.7 MP and the D700 12 MP.

Unfortunately those MP values are not comparable between the two technologies.

But - they are the best numbers for each technology.

And - as has been shown here zillions of times - there do not exist a reasonable common measure. Maybe you should divide the Bayer CFA value with 2 - but that has no technical motivation - only marketing. Instead Foveon multiplies by 3. That has both technical and marketing motivation - but is (IMHO) totally useless. You could count transistors or diodes instead; just as interesting.

--
Roland

 

[Roland Karlsson]

Hi! So you are the one who owns both SD14 & K20d,

I think he meant Canon 10D and 20D.

They are older 6 and 8 MP cameras, the 10D is as old as Sigma SD10.

But they were good cameras - so it is interesting to know his opinion there also.

--
Roland

 

[gbainbridge]

Maybe you should divide the Bayer CFA
value with 2 - but that has no technical motivation - only marketing.

I disagree totally. It does have a technical motiviation as the Beyer chip can never reproduce any more luminance detail than the number of green pixels.

Instead Foveon multiplies by 3. That has both technical and marketing
motivation

Unfortunately that has no technical motivation, purely marketing. Sigma (and CIPA) only make that claim as it compares well with the CFA claims, but it's nonsense as the Sigmas cannot produce a 14 Mpixel image without interpolation any more than the CFAs can. Like it or not the SD14 is a 4.7 megapixel camera.

--
Thanks,
Gary.

 

[Erik Magnuson]

I disagree totally. It does have a technical motiviation as the Beyer
chip can never reproduce any more luminance detail than the number of
green pixels.

Since almost all CFA cameras do reproduce more detail than this, I don't know why you say this. They typically produce resolution equivalent to 70% (JPEG) to over 80% (raw with aggressive detail recovery) of the total pixel count. There is a reason the green pixels are on diagonals.

--
Erik

 

[Ed_S]

Roland and all,

I kind of agree with your (Roland's) last and detailed comments to me. Distilled to its essence (for me), the metrics of each system are what they are. Trying to equate MPs between systems is somewhat subjective and IMHO somewhat situational. As we've discussed at great length over the months and years, there are situations where sometimes measurably other times subjectively one system excels over another and seems to provide results in excess of - or closer to - what might be expected based upon pure pixel counts.

Unfortunately - and hence the marketing component - many if not most mass market customers (and sales types) do not have the interest or technical depth to explain/understand each and draw informed conclusions. I'm not trying to be elitist - heaven knows I don't have the technical elan to be an "expert" in all this. But MP has been entrenched as the be all end all yardstick for such discussions in the general media and marketplace and hence, the pixel wars ensue.

Kind regards,
--
Ed_S
http://www.pbase.com/ecsquires

 

[Roland Karlsson]

Maybe you should divide the Bayer CFA
value with 2 - but that has no technical motivation - only marketing.

I disagree totally. It does have a technical motiviation as the Beyer
chip can never reproduce any more luminance detail than the number of
green pixels.

Ahhh ... I should be more careful before being so sure about that.

It is a myth IMNSHO that you can only use the green detectors for luminosity. You can be more clever than that - a dangerous choice as you may be inventing more details than there really are in the image. But it is possible - and it is done - and (as far as I know) with some success.

Instead Foveon multiplies by 3. That has both technical and marketing
motivation

Unfortunately that has no technical motivation, purely marketing.
Sigma (and CIPA) only make that claim as it compares well with the
CFA claims, but it's nonsense as the Sigmas cannot produce a 14
Mpixel image without interpolation any more than the CFAs can. Like
it or not the SD14 is a 4.7 megapixel camera.

I tend to agree with you here.

But I know that lots of people disagree. And they have a point IMHO. There are 14 MP in the SD14 sensor - 4.7 of each color. A play with words maybe. But (unfortunately) the pixel war is a play with words - so why not?

--
Roland

 

[Zone8]

When Hyam A. Photographer went to heaven, he got a bit bored after the first hundred years so he asked God to give him something that would occupy him for ever.

So God set him the task of counting pixels on all the sensors known to man.

3,000 years later, he had finished, albeit with some arguments along the way of course.

So, he again asked God to give him a task that really would keep him occupied for ever.

So God said to him:

"Go back down to Earth and go to the Three Tuns pub on Dartmoor and wait for the landlord to buy you a drink".

(That's an adopted form of a copperplate written similar tale hanging framed behind a certain pub in a small village on Dartmoor in Devon in the UK - but I think you will (well, as photographers?) get the picture!

--
Zone8

The photograph isolates and perpetuates a moment of time: an important and revealing moment, or an unimportant and meaningless one, depending upon the photographer's understanding of his subject and mastery of his process. -Edward Weston
http://www.photosnowdonia.co.uk/ZPS

 

[gbainbridge]

Since almost all CFA cameras do reproduce more detail than this, I
don't know why you say this.

I would like to see proof of this. It's certainly not my experience, and I do have two CFA cameras, albeit not SLRs.

It may be possible to simulate a greater level of detail by, for example, using clever interpolation and sharpening techniques, but I cannot see that a CFA sensor can ever correctly detect a piece of detail which, without the AA filter, would illuminate a single pixel.

--
Thanks,
Gary.