Super CCD HR pixel doubling for greater resolution

In a lot of past reading of this or that photographic information-source I picked up the idea that a Fuji Super CCD HR sensor can provide a bit of extra resolution via "pixel-doubling" - using a RAW processor to render an image with twice the number of pixels that is obtained from the native resolution of the sensor. As I recall, Fuji themselves claimed in the early days that a camera was, say, 6Mp resolution rather than the mere 3Mp of actual photo-sensors in the camera.

Many authoritative sources agreed with Fuji to a point but added caveats. Typically they would say that the resolution was not doubled but that some extra information - extra resolution - could be got out of the sensor because the extra pixels were generated from some clever recombination of the individual photo-diode data, a process only possible because of the physical arrangement of the Super CCD HR sensor's octagonal photo-diodes.

Encouraged by all this talk I did, when developing RAW files from a Fuji S9500, double the pixels from 9 to 18Mp in ACR. Subsequent development down to the eventual 18Mp jpeg image did seem to smooth out the jagg of diagonal lines and allow a certain "finer" look to the pictures. I developed a dozen or so of the same RAW files as both a 9Mp and an 18Mp image to do direct comparisons.

But.....my wee experiment was hardly scientific and relied on my subjective impressions of the images.

Now I have a Fuji S100FS with the last (I understand) version of a Super CCD HR sensor in it. The same (in this case) 11Mp to 22Mp image-size comparisons do seem to show that the larger sized images are a little more "fine".............?

Does anyone out there have information of a less subjective kind that demonstrates if this pixel-doubling RAW development process really does provide measurable increase in resolution - the sort measured with a test chart? If so I'd be grateful for a link or perhaps even a summary.

SirLataxe, not too keen on the huge file-sizes of pixel-doubled images but still a resolution-junky.

How's this shot for resolution, with a Fuji S2 Pro and the low-priced Tamron 100-300, at the long end?

You're right, except it actually work with SR (S pro serie) sensors. Not HR sensors, since those don't have dual photodetectors.
--
http://www.flickr.com/photos/maximebrousse/

Maxime,

Yes, I understand that a Super CCD SR sensor has the two kinds/sizes of photo-diodes (in pairs) which are used in combination to give a greater dynamic range (at N pixel resolution) or individually to give more pixels (2 X N pixel resolution).

However, the stuff I read about getting more resolution from pixel-doubling did refer to the Super CCD HR sensor. Although this sensor has only one type of photo-diode they are all arranged in a different matrix from the more usual sensor and are also octagonal rather than rectilinear. The image-pixels can be generated in such a way that twice as many as the photo-diode-count may "interpolated" from different pairings of the photo-diodes within the matrix.

Well, this is what the Fuji blurb claimed at the time the HR sensor was announced. I don't pretend to understand exactly what happens within the technologicals myself, mind. Here is a bit of a summary from Wikipedia:

http://en.wikipedia.org/wiki/Super_CCD

So, my question remains - does pixel-doubling really provide a bit more resolution, even if it isn't really the same as having twice as many photo-diodes? Is anyone aware of a test done to measure test-chart resolution from a doubled-pixel picture against the same picture made from non-doubled-pixels?

SirLataxe, interested more in the effects than the causes, in this instance.

If i remember correctly, DPReview had reviewed some older superccd cams and said that it does increase resolution, not by double, more like 1.5X. As a drawback, it made strange visual artifacts in the image due to the interpolation. You might want to read some older fuji camera reviews and it gives you a camparison with other similar cameras.

SirLataxe wrote:

In a lot of past reading of this or that photographic information-source I picked up the idea that a Fuji Super CCD HR sensor can provide a bit of extra resolution via "pixel-doubling" - using a RAW processor to render an image with twice the number of pixels that is obtained from the native resolution of the sensor. As I recall, Fuji themselves claimed in the early days that a camera was, say, 6Mp resolution rather than the mere 3Mp of actual photo-sensors in the camera.

Many authoritative sources agreed with Fuji to a point but added caveats. Typically they would say that the resolution was not doubled but that some extra information - extra resolution - could be got out of the sensor because the extra pixels were generated from some clever recombination of the individual photo-diode data, a process only possible because of the physical arrangement of the Super CCD HR sensor's octagonal photo-diodes.

. . .

Does anyone out there have information of a less subjective kind that demonstrates if this pixel-doubling RAW development process really does provide measurable increase in resolution - the sort measured with a test chart? If so I'd be grateful for a link or perhaps even a summary.

One of the caveats that I recall was that while there was only a very slight actual resolution increase from the doubled pixel images, the increase seemed more dramatic than actual when looking at most resolution test charts that at the time primarily relied on horizontal and vertical lines, which are also more prevalent in real life. For more random patterns such as leaves, no big difference. It just so happens that I read DPR's S5100 review a day or two ago (I bought one in 2004) and it had this to say about the 3mp S5000 and the 4mp S5100 :

Excellent resolution - amongst the best results we've seen from any of the super zoom cameras, even those, such as the Kodak DX7590, with higher pixel counts. Perhaps not surprisingly, the results are considerably better than the S5500/S5100's predecessor (the Super CCD-equipped S5000, which boasts an interpolated output of 6 megapixels).

http://www.dpreview.com/reviews/fujifilms5100/page8.asp

So there you have it. If the S5000's 6mp images are better than its 3mp images, it's not by enough to match the S5100's 4mp images.

SirLataxe, not too keen on the huge file-sizes of pixel-doubled images but still a resolution-junky.

The S100fs's 11mp images produce impressively high resolution images, but based on DPR's Studio Scene crops, Canon's G10 produces higher resolution images at base ISO, but the S100fs does slightly better at ISO 400 much better at ISO 1600.

Fuji's diagonal sensor designs in these cameras forces ALL images to be interpolated up tp twice the size in order to rotate the image back to a normal orientation (rotopolation). So does it really matter if you let the camera do it with the jpg engine, or do it yourself with a RAW file? I recall reading many articles on this back in the day and it seemed as if most experiments showed the resolution was better when using these sensors at their highest resolution setting (or RAW).

Tests of the S602 which was a 3 megapixel chip which could output a 6 megapixel file showed resolution on par with a 4 to 4.5 megapixel camera when used in 6 megapixel mode. I think anytime you use RAW your giving yourselfore to work with just by dint of the fact you haven't let Fuji;s noise reduction algorithm loose on the image.
Ted

Ted

SirLataxe wrote:

Maxime,

Yes, I understand that a Super CCD SR sensor has the two kinds/sizes of photo-diodes (in pairs) which are used in combination to give a greater dynamic range (at N pixel resolution) or individually to give more pixels (2 X N pixel resolution).

However, the stuff I read about getting more resolution from pixel-doubling did refer to the Super CCD HR sensor. Although this sensor has only one type of photo-diode they are all arranged in a different matrix from the more usual sensor and are also octagonal rather than rectilinear. The image-pixels can be generated in such a way that twice as many as the photo-diode-count may "interpolated" from different pairings of the photo-diodes within the matrix.

Well, this is what the Fuji blurb claimed at the time the HR sensor was announced. I don't pretend to understand exactly what happens within the technologicals myself, mind. Here is a bit of a summary from Wikipedia:

http://en.wikipedia.org/wiki/Super_CCD

So, my question remains - does pixel-doubling really provide a bit more resolution, even if it isn't really the same as having twice as many photo-diodes? Is anyone aware of a test done to measure test-chart resolution from a doubled-pixel picture against the same picture made from non-doubled-pixels?

SirLataxe, interested more in the effects than the causes, in this instance.

Kim,

You make interesting observations but I'm not entirely convinced by your arguments.

First, another question: does the demosacing algo of ACR mimic that of the Fuji RAW converter? If so, is it not an advantage to allow ACR to generate the 2 X pixel count (22Mp in the case of the S100FS)?

As you say, the extra resolution obtained is small and probably to do with the smoothing out of otherwise jagged diagonals (due to the physical arrangement of the photo-diodes in the HR sensor), When I examine the 22Mp images at 100% or more, such smoothing of jagged diagonals is unmistakeable.

Secondly, I question the mantra that "interpolation creates nothing". It may not create data obtained from "the real" but it can and does create data that attempts to guess the real. Sometimes it guesses very well.

An obvious example - using Photoshop bicubic-smooth to upsize the number of image pixels creates pixels that fit nicely between those from the real world. If it didn't, the upsized picture would be full of random noise. Of course, the degree to which the guessed-at interpolated pixels are accurate guesses of "the real" gets worse as the amount of interpolation increases.

Finally, I think your dismissal of the need to test for resolution improvements goes agin' a very basic principle of the rational approach to these matters. Theories and logic-chopping are very useful but in the final analysis, reality wins. A test that looks at reality will always trump the logical conclusion of a theory. Of course, it is theories that suggest what to test and how.........

SirLataxe, still looking for a definitive test chart examination of Super CCD HR sensors, then.