SD9 samples - extreme images!

There is a good reason why the SD-9 tops out at ISO 400, and why the images are noisier than their equivalent D60 images at the equivalent ISO. The reason is that the SD-9 has approximately 3x the number of sensors, and as such, generates approximately 3x the amount of noise.

Think of it this way. For each pixel, the SD-9 has to generate 3 seperate values from three seperate photo-sites. Given that each has the chance of producinga noise value equivalent to a sensor of equal size from say a D60, since they have to combine values to generate a single color value, there is about a 3x spread on the chance of 1 of the three sensors contributing a "noise" value and thus throwing the entire pixel off.

Granted, unless all three pixels generate a noise value, the noise will be less on a per pixel basis (i.e. smoother noise), but let's face it, noise is noise, and it will be apparent.

This is also the reason for the decrease in ISO to approximately 1/3 the value one would expect from such a camera. This is an inherent problem with the Foveon design, which I expect will become less pronounced (but always present) as all CMOS designs increase in their light gathering capability.

This is a fine camera for certain work, but these limitations will keep it from being used for anything other than controlled situations.

Peter

Zoltan Csuka said:

Phil Askey said:

Having handled and shot with the prototype kit last year I'm now
used to it, but it may need some explaining in my review.

Anders Lundholm said:

Hello everbody!

I've been looking at the sample images posted from the SD9 camera
for half an hour now. I have to admit that my expectations from
this sensor are high, but the samples look almost computer
generated (that is, extremely crisp).

Crisp edges, crisp colors .. I'm amazed. I'm truly amazed. I never
ever thought the X3 sensor would bring such image quality to
digital photography. The samples posted are by far the best output
I've seen from any DSLR camera so far.

I'm sure Foveon is expected to announce higher resolution X3
sensors in the near future.

We all want high resolution, superb image quality and color
fidelity from our cameras, but I'm not sure how e.g. Canon or Nikon
are going to react to this new sensor.

Another thead discusses the fact that japanese coorporations tend
to keep their investments inside Japan .. exiting!

--
with regards
anders lundholm · [email protected]
the sphereworx / monoliner experience

Click to expand...

--
Phil Askey
Editor / Owner, dpreview.com

Click to expand...

--
Best wishes,
Zoli

Click to expand...

--
Peter Sills
Digital Focus
http://www.digitalfocus.net

Peter Sills said:

There is a good reason why the SD-9 tops out at ISO 400, and why
the images are noisier than their equivalent D60 images at the
equivalent ISO. The reason is that the SD-9 has approximately 3x
the number of sensors, and as such, generates approximately 3x the
amount of noise.

Click to expand...

Having more sensors would reduce the appearance of noise, not increase it.

Member said:

Think of it this way. For each pixel, the SD-9 has to generate 3
seperate values from three seperate photo-sites. Given that each
has the chance of producinga noise value equivalent to a sensor of
equal size from say a D60, since they have to combine values to
generate a single color value, there is about a 3x spread on the
chance of 1 of the three sensors contributing a "noise" value and
thus throwing the entire pixel off.

Click to expand...

There is probability 1 that there will be noise in any pixel arrangement. If you have more sensors, you have a better chance that the noise will be averaged out.

Member said:

This is also the reason for the decrease in ISO to approximately
1/3 the value one would expect from such a camera. This is an
inherent problem with the Foveon design, which I expect will become
less pronounced (but always present) as all CMOS designs increase
in their light gathering capability.

Click to expand...

I think a better explanation is that the chips has 3X as many transistors per pixel as a Bayer pattern sensor. The transistors block light.

Also, if we look at the diagrams in the patent, it appears that the top photodiodes are smaller than the bottom ones. If this is an accurate representation of what's on the chip, then the sensitivity may be limited by the smallest of these three layers.

--
Ron Parr
FAQ: http://www.cs.duke.edu/~parr/photography/faq.html
Gallery: http://www.pbase.com/parr/

Ron Parr said:

Peter Sills said:

There is a good reason why the SD-9 tops out at ISO 400, and why
the images are noisier than their equivalent D60 images at the
equivalent ISO. The reason is that the SD-9 has approximately 3x
the number of sensors, and as such, generates approximately 3x the
amount of noise.

Click to expand...

Having more sensors would reduce the appearance of noise, not
increase it.

Click to expand...

This would be true if each of these sensors was equivalent to each other. They are not. They each contribute to 1 pixel. So each pixel has a 3x increase in chance of being affected to some degree.

Member said:

Member said:

Think of it this way. For each pixel, the SD-9 has to generate 3
seperate values from three seperate photo-sites. Given that each
has the chance of producinga noise value equivalent to a sensor of
equal size from say a D60, since they have to combine values to
generate a single color value, there is about a 3x spread on the
chance of 1 of the three sensors contributing a "noise" value and
thus throwing the entire pixel off.

Click to expand...

There is probability 1 that there will be noise in any pixel
arrangement. If you have more sensors, you have a better chance
that the noise will be averaged out.

Click to expand...

Again, not true as the sensors are not equal, ie they are not "averaged" they are summed to make up the final pixel, each contributes one component to the color. Ergo any noise in the individual sensor will contribute noise to the final image. Additionallyl, each of these sensors must be smaller than a single sensor in an equivalent mpixel Bayer design. As we know, smaller sensors are much more prone to noise artifacts.

Member said:

Member said:

This is also the reason for the decrease in ISO to approximately
1/3 the value one would expect from such a camera. This is an
inherent problem with the Foveon design, which I expect will become
less pronounced (but always present) as all CMOS designs increase
in their light gathering capability.

Click to expand...

I think a better explanation is that the chips has 3X as many
transistors per pixel as a Bayer pattern sensor. The transistors
block light.

Click to expand...

If this were true it would be disproportionate. The sensors on the top layer would get much more light than on the bottom. While I am sure this is true to some degree, I am sure it is insignificant, or it would cause more problems than just noise.

Member said:

Also, if we look at the diagrams in the patent, it appears that the
top photodiodes are smaller than the bottom ones. If this is an
accurate representation of what's on the chip, then the sensitivity
may be limited by the smallest of these three layers.

--
Ron Parr
FAQ: http://www.cs.duke.edu/~parr/photography/faq.html
Gallery: http://www.pbase.com/parr/

Click to expand...

--
Peter Sills
Digital Focus
http://www.digitalfocus.net

Phil Askey said:

I feel that there is also a certain amount of concern from myself
and Foveon that people aren't READY to see this amount of
sharpness, you're right because we're all used to looking at Bayer
generated images at 200% / 300%, when you do so to a Foveon image
it looks like nothing you've seen before, crisp sharp boundaries
and very 'computer like'... But of course it's actually just
single pixel resolution that we're not used to.

Click to expand...

That's pretty wild. I don't usually look at images zoomed in more than 100% but after reading this I took a zoomed in look at one of the TIFF samples.

You're right, that is very different and 'computer like'--very unlike images from other digital cameras or scanned film. I'm not sure I like it, but what it looks like zoomed in that far isn't as important as how it looks printed and/or viewed at 'normal' size.

Interesting.

Peter Sills said:

Member said:

Having more sensors would reduce the appearance of noise, not
increase it.

Click to expand...

This would be true if each of these sensors was equivalent to each
other. They are not. They each contribute to 1 pixel. So each
pixel has a 3x increase in chance of being affected to some degree.

Click to expand...

No. Your reasoning is just plain wrong, but if we follow it to the (illogical) conclusion we see that things would be even worse in a Bayer pattern sensor since interpolation algorithm considers values from 9 (often more) neighboring pixels, so the values would be even more likely to be off using your reasoning.

Member said:

Member said:

There is probability 1 that there will be noise in any pixel
arrangement. If you have more sensors, you have a better chance
that the noise will be averaged out.

Click to expand...

Again, not true as the sensors are not equal, ie they are not
"averaged" they are summed to make up the final pixel, each
contributes one component to the color.

Click to expand...

And how many pixels contribute in a Bayer pattern sensor?

Member said:

Ergo any noise in the
individual sensor will contribute noise to the final image.
Additionallyl, each of these sensors must be smaller than a single
sensor in an equivalent mpixel Bayer design. As we know, smaller
sensors are much more prone to noise artifacts.

Click to expand...

Nope. While each photodiode is thinner in the Z dimension it is precisely because it is detecting the photons at the depth at which they are absorbed. Consider a traditional photodiode in a Bayer pattern sensor with a red filter on top of it. It's true that the filter is deeper, but the extra depth would just detect non-red wavelength photons, which are blocked by the filter anyway.

Member said:

If this were true it would be disproportionate. The sensors on the
top layer would get much more light than on the bottom. While I am
sure this is true to some degree, I am sure it is insignificant, or
it would cause more problems than just noise.

Click to expand...

It's a fact of every sensor that some colors are more sensitive than others. This is because the substrate is more sensitive to some wavelengths than others. Even Bayer patterns sensors much correct for this.

--
Ron Parr
FAQ: http://www.cs.duke.edu/~parr/photography/faq.html
Gallery: http://www.pbase.com/parr/

Hope you're having fun at the show.

The post I was replying to suggested that further sharpening of the images isn't possible. My suggestions were just meant to show that there are things you can do to the image (e.g., edge enhancement) to make elements stand out and create useful and hopefully salable effects, but that treating these images like Bayer images will not give the desired results.

Phil Askey said:

That's because most of the things you speak of were designed to fix
the sharpness problems inherent in Bayer CCD / (some) scanned
images.

Click to expand...

I have zoomed in 300 to 400% on both the tiff and jpgs and see no sharpening halos whatsoever.

--
http://www.pbase.com/galleries/sasc

I don't understand all the technical stuff but I do know what looks good to my eye and these foveon pictures look great. They can be upsampled and sharpened and jpged and they still look great. I have an older sample from Foveon site of a cat. I take a lot of cat and feathered critter shots and the jaggies that my D30 shows on whiskers and feathers is very aggravating. The foveon cat can be sharpened, upsampled, downsampled and jpg compressed and it shows no jaggies whatsoever. And the detail and sharpness in the leaf shots is amazing.

Now if I could afford to buy one and new lenses I would.

SC said:

Now if I could afford to buy one and new lenses I would.

Click to expand...

Agreed. If I could afford to be out on the bleeding edge of technology, this is definitely one camera I would consider. It will definitely require a "different way" of looking at and working with digital images. One of these days I will probably get a quality digital SLR. Since I don't expect to do so however for at least a year, I will get to enjoy watching the march of progress. I will definitely keep my eye on the Foveon technology. Glad it is finally here.

--
Photography by David 'Eaglechild' Robinson.
My Photos: http://www.pbase.com/brdavid/
Nikon 5700 forum web site : http://www.eagles-nest.org/teameagle/