SD14 JPEG interpolation

[Jay Turberville]

I agree with this thinking, which is why though Sigma needs a
double-size JPG poutput to refocus the conversation and let people
be a little dissapinted it's not quite 14MP instead of wondering
why you'd buy a 4.7MP camera.

But they should have simply doubled the photosite count and gone with a 9Mp or 10Mp Jpeg. That should compare very favorably with the current 10Mp CFA offerings on an image pixel per image pixel basis and drivien home the point about pixel count and their design.

--
Jay Turberville
http://www.jayandwanda.com

 

[Jay Turberville]

It has to be one number ... and it has to be inaccurate. To make it
more complex or exact is not doable.

Yep. You probably want to say it something like "In a class similar to 10Mp CFA cameras."

--
Jay Turberville
http://www.jayandwanda.com

 

[paulbod3]

Reverse thinking :

the position of a point in the plane can be unequivocally identified by triangulation.

In RGB, if RGB is stacked, we need 3 pix x3, if not (bayer) 9 pix, 3 each RGB, irregularly spaced.

Color depht of this point is univocal in Foveon, as well as in bayer, so we are using the same amount of pixels.

If we think of the larger Jpeg not as an expansion of the Raw size, but as a larger frame where we are going to place the positions of the smaller Raw, our precision in doing so is higher obviously with Foveon, as long as the information needed doesn't exceed what is available in Raw.

So for simpler pictures, portraits etc, even the double mode could be at an advantage against bayer for the same count (3.4x3 vs 10.2) even for prints over A4 (7.7mp at 314ppi).

When the information needed raises, say a very detailed urban landscape, then bayer may lead (10.2 vs 3.4x3)because the spatial info needed is higher.

IMHO there is no simple way to compare the resolution of the two systems.

10.2mp bayer (2.55R 5.1G 2.55B) is "perfect" if resized at 2.55mp,
10.2mp foveon (3.4 each RGB) is "perfect" at 3.4
14mp foveon (4.65 each RGB) is "perfect" at 4.65

Over such sizes, interpolation is needed, and bayer can be b+w "perfect" at 10.2mp, but B+W !

 

[Jay Turberville]

-- I picked this image for some sharp, angled edges taken with the
sharp 50mm prime in harsh winter afternoon sunlight, and the
advantage Jay mentioned does show. There isn't much failure of
aliasing, but it is smoothed to invisibility in the 1.4x version.

Actually, it isn't smoothed so much as it is no longer being misrepresented to the same degree. A pixel isn't a little square. It is more like a tiny little "blob".

So - this 14 mp output choice may be as much a choice of the
engineering team as any marketing.

Except that 1.4x yields a 9Mp image. I think a 10Mp image would have been about right.

--
Jay Turberville
http://www.jayandwanda.com

 

[Kendall Helmstetter Gelner]

Because 14M hints at it being the number of detectors that is
counted ... and we are back on square zero again. 4.7M (classic)
pixels or 10M Equiv pixels are the only ones I think is OK.

Yes, it's the same square but everyone's facing a different direction!

So - whats your propsed range for the SD14?

Probably 9-11MP, but I'd have to see some comparison shots to be more sure of that...

Yes, your number also happens to fall right in the middle of my range but that doesn't mean I can make use of it any easier!

I fully agree. But a number is unfortunately neccessary.

A range works just as well though really - I use a simlarily derived range (6-8MP) when describing the output of the SD-10 to people, along with a disclaimer that the sesnor is very different which is why I can't give them one number.

And unfortunately ... now it is half past midnight in Sweden and it
is bed time!

Good night, and thank you for a pleasant discussion.

--
---> Kendall
http://InsideAperture.com
http://www.pbase.com/kgelner
http://www.pbase.com/sigmasd9/user_home

 

[Kendall Helmstetter Gelner]

This is very interesting, and I don't think it is just my
eyeglasses. Birds, antennas, and the intricate white and beige
inlaid tiling of an image on the side of our Rathus (city hall) all
are much more present in the picture.

Another point for this I would make is that many people are used to looking at Bayer images, even up close. If you are used to this Foveon images do look smaller, although very sharp....

It's only when you upsize them to some degree that you are "looking" at the same kinds of pixels, basically pixels with some interpolation between them.

It's also a difference if you use one exact scale number generally to examine image detail, say 100%. If you are used to zooming into a particular scale for review again the Foveon image will benefit from some upscaling because it will "look" larger and help you see the detail captured at that given percentage a bit more.

And you're right that it smooths the whole thing out a little more so any jaggies mellow out a little and cease to detract fom close review.

By the way, Kendall, I appreciate that you talk to Roland. It is
very clearly what he wants.

Err, I'm not quite sure what you mean either but I had a very nice set of conversations today so I was happy...

--
---> Kendall
http://InsideAperture.com
http://www.pbase.com/kgelner
http://www.pbase.com/sigmasd9/user_home

 

[NarrBL]

Another point for this I would make is that many people are used to
looking at Bayer images, even up close. If you are used to this
Foveon images do look smaller, although very sharp....

Yes, exactly. I didn't feel to make such points so directly - feel it's better when people arrive so themselves. But there's an interesting doorway here, it does seem.

A nice evening when it gets to you there, Kendall.

Clive

 

[ole thofte]

and for you, Kendall.

After all this endless noise, I decided to try a 1.4 Bicubic upsize
on a SD10 image of my own.

The result: something surprising.

-- It now has many qualities a great deal like the 'best' images
from a Canon 1ds, in its 'look' at full size. It is a bit sharper.

-- many details that are present but subtle at 1x become much more
visible at 1.4x.

That was an interesting idea Clive and I tried it out on a random shot of mine. It really is surprising and better than I thought. What a lot of details in a picture like that....
Here's my experiment 1,4x http://www.pbase.com/image/67639705

Ole
--
http://www.pbase.com/thofte

 

[JohnClif]

We all know that Bayer cameras typically show about 62% of the theoretical resolution (if we don't, that's a factoid I've seen based upon demosaicing). Blame the loss to a combination of demosaicing and anti-aliasing (nutella).

IF... the SD14 can produce a 14 MP JPEG that is comparable in resolved detail to, say, a Canon 5D, then why wouldn't people buy an SD14 for half the price?

Question is, CAN the SD14 produce a 14 MP file with as good, or better, resolved detail than that from a Canon 5D?

Based upon the performance of the SD10 against the 5D on a pixel-by-pixel basis, and on the fact that the SD14's photosite size gives it about 20% more resolving ability in terms of lines/mm, I think there's a good chance that the answer is, "Yes!"

If so, if the 14 MP JPEG from the SD14 compares very favorably against the 12 MP JPEG from a Canon 5D, then the new Sigma dSLR will be in great demand.

--
'Do you think a man can change his destiny?'
'I think a man does what he can until his destiny is revealed.'

 

[Just Looking]

That was an interesting idea Clive and I tried it out on a random
shot of mine. It really is surprising and better than I thought.
What a lot of details in a picture like that....
Here's my experiment 1,4x http://www.pbase.com/image/67639705

Did you start with one that was overly sharpened already? The flowers have funny halos around them.

j

 

[ole thofte]

That was an interesting idea Clive and I tried it out on a random
shot of mine. It really is surprising and better than I thought.
What a lot of details in a picture like that....
Here's my experiment 1,4x http://www.pbase.com/image/67639705

Did you start with one that was overly sharpened already? The
flowers have funny halos around them.

JL, you're right. I didn't see that, but when I used a magnifying glass it was quite plain. I wish I knew I used Lightroom (macbook) and developed from raw using BobNL's standard....but maybe I added 25% sharpness. I put up another one using Lightroom's flat developing without any sharpness. It doesn't have halos as far as I can see, but the colors are 'flat'.

On the jazz photo I added 25% sharpness before reesharpening, but there's no halos as far as I can see?

Ole
--
http://www.pbase.com/thofte

 

[ole thofte]

...resizing instead of resharpening.
--
http://www.pbase.com/thofte

 

[Jay Turberville]

Question is, CAN the SD14 produce a 14 MP file with as good, or
better, resolved detail than that from a Canon 5D?

Because purchasing a 24x36mm camera often involves issues other than image base resolution/quality issues. Lens options, DoF ranges and viewfinder experience are things I often hear cited.

Based upon the performance of the SD10 against the 5D on a
pixel-by-pixel basis, and on the fact that the SD14's photosite
size gives it about 20% more resolving ability in terms of
lines/mm, I think there's a good chance that the answer is, "Yes!"

We never settle the "equivalency" issue to everyone's satisfaction with the SD10. I doubt we will with the SD-14 either.

If so, if the 14 MP JPEG from the SD14 compares very favorably
against the 12 MP JPEG from a Canon 5D, then the new Sigma dSLR
will be in great demand.

Nope. It doesn't necessarily follow. There is more to the issue that base image quality.

--
Jay Turberville
http://www.jayandwanda.com

 

[BarrytheB]

very interesting discourse, I enjoyed the observations and also the civility- maybe Phil will come to some meaningful conclusions in his side by side comparisons.
--
Barry Byrd
http://www.pbase.com/barryb
http://www.pbase.com/sigmadslr

 

[Edwin M]

Actually, it seems that Foveon has done something quite brilliant- they have used bayer to beat bayer! Since they are competing against mosaic imagers, they don't advertise this, but it's in their patent. The new chip has technology that interpolates the 4.6 mp 3d array into a 14mp 3d array using what would amount to an "inverse bayer algorithm". Apparently, foveon got to thinking, "hmm... since sophisticated bayer algorithms can make beautiful images out of scarce raw data, what would they do for our rich raw data?" SPP 2.1 uses bicubic to double both the horizontal and vertical pixel count. If this were used on the 4.66MP raw sensor data, we would arrive at a 18.75MP upscaled image. This is strong evidence that bicubic interpolation is not being used in the new sensor. If however foveon uses the bayer like intelligent upscaling that is indicated in the patent, we would arrive at the 14mp upscaled image. This intelligent upscaling is superior to bicubic and should produce images that compete with 16mp cameras. Of course, it's possible that their patented technology didn't make it into the chip at all, and their just doing a partial bicubic upscale!

 

[Roland Karlsson]

So - whats your propsed range for the SD14?

Probably 9-11MP, but I'd have to see some comparison shots to be
more sure of that...

Yes, your number also happens to fall right in the middle of my
range but that doesn't mean I can make use of it any easier!

9-11 is OK. Its just 20% difference and there will not be heated discussion regarding if 9 or 11 is OK. Enough people is mature enough for avoiding that, I hope

--
Roland
http://klotjohan.mine.nu/~roland/

 

[Roland Karlsson]

...

It looks quite sure that no such functionality is used in the new chip. So ... nope ... this is not it IMHO. It was discussed earlier in this forum, before Fotokina. And I have heard that it "irritated" Foveon.

--
Roland
http://klotjohan.mine.nu/~roland/

 

[SergeiK]

I suspect the 14Mp images are intented exclusively for camera reviewers such as dpreview.com or customers who will compare images themselves.
They get a possibility to directly compare resolution without upsizing.

 

[To shoot or]

According to Sigma UK SPP 3.0 will interpolate half or double the original pixel amount to max 18,58 Mega pixel.

Source: http://www.sigma-imaging-uk.com/cameras/sd14/spp3.htm

"SIGMA Photo Pro 3.0 supports 8-bit TIFF, 16-bit TIFF, and JPEG file formats and also outputs images in sRGB, Apple RGB, Adobe RGB and ColourMatch RGB colour spaces. The default setting for resolution is the same as the resolution selected when capturing the image on the camera (High, Medium, or Low). It is also possible to half, (for sending via e-mails) or double (for high quality enlargements, max. 18.58 million pixels:5,280×3,520pixels) the resolution when processing the image."

That leaves us with the the question: 14 MP in camera vs 18,6 MP in SPP3.0 or multiply factor 1,75X vs a plain 2x ...

To shoot or ...

 

[Ante Vukorepa]

Pardon my french, but - boll* ks!
It is an overrepresentation, but in color resolution.

Besides, things aren't as grim with Bayer as they are usually portrayed. New Bayer interpolation algorithms (like adaptive homogeneity-directed demosaicing algorithm) have gone leaps and bounds over the plain vanilla Bayer demosaic.

Here's an example:



This is a 100% crop of a 13MP EOS 5D file, processed from RAW via DCRAW. Now try to tell me this doesn't have as much spatial resolution as advertised.

It's the Bayer artifacts (color in particular) people should be using to scare their children before bedtime, not resolution. Sorry.

It is bayer that is the over representation. A 10MP bayer camera
will typically be equivalent to 5MP in they eyes of human users,
but under certain conditions could fall as low as 2.5MP (but this
conditions are mostly artificial). How they heck do you accurately
represent that in a number. I use Bayer divided by two as a general
case.

--
--------------------------------------------
Ante Vukorepa