Pages: [1]   Go Down

Author Topic: Puzzled by Fuji S2 Interpolation  (Read 3076 times)

mbridgers

  • Full Member
  • ***
  • Offline Offline
  • Posts: 156
Puzzled by Fuji S2 Interpolation
« on: January 16, 2003, 10:36:58 am »

This is really a can of worms.  Fuji claim that their Super CCD technology, by using octagonal photosites vs. square, allows for more resolution.  It is my understanding that all digital pictures are processed in the camera, either for color (standard CCD) or overall pictures.  DPReview has an older article describing all this.  DP Review article

Bottom line, of course, is image quality, and the S2 seems to deliver in that department, judging from samples I've seen.
Logged

Dale_Cotton

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 588
    • http://daystarvisions.com
Puzzled by Fuji S2 Interpolation
« Reply #1 on: January 16, 2003, 09:01:44 pm »

One important thing about the octagonal pattern is that each photosite is closer its neighbours than in other sensors. And as Robert says the fact that the pixels do not fall into the same grid pattern as they have to end up in when in an image file, something has to happen. Because of the photosite density the 6 million pixels in the S2's camera do yield a real gain in resolution. You can see this quite objectively in looking at the results of the ISO 12233 test target numbers on dpreview.com and imaging-resource.com. After studying many sample image my own guesstimate is that the S2's sensor has a resolution equivalent to somewhere between 7 or 8 megapixels in a conventional array. Whether there is a way to convert from the S2's raw output to a 7 or 8 million pixel image file with no loss is another question, however.

What is clear is that the 6 million pixel output file has lower resolution on the USAF target test, so data are being lost. People who presume that, because the camera has a 6 megapixel count, the 6 million pixel image file must therefore be the full raw data are simply mistaken. We see on the bottom of this page in Phil Askey's review that the S2 in 12 mp mode weighs in at 1800 horizontal and 1700 vertical, but only at 1650 and 1400 in 6 mp mode.
Logged

Cliff LeSergent

  • Newbie
  • *
  • Offline Offline
  • Posts: 14
    • http://www.cl-photos.com
Puzzled by Fuji S2 Interpolation
« Reply #2 on: February 03, 2003, 11:44:30 am »

Thanks for your responses, but I should point out that the question does not directly concern the camera's resolution, nor am I trying to start a debate as to the camera's quality;  I'm merely trying to satisfy my technical curiosity.

Interpolating a file does not increase its resolution, only its size.  Since (from what I've been told) apparently the S2 is only capable of producing interpolated images in RAW/TIFF mode, the result is an image file with decreased resolution and increased size.  What happens when you want a smaller file size?  How does the camera decide which pixels to discard?  

I guess what I'm asking is if the camera has some sort of statistical algorithm which prevents it from further degrading the image when you do these subsequent interpolations.
Logged

Cliff LeSergent

  • Newbie
  • *
  • Offline Offline
  • Posts: 14
    • http://www.cl-photos.com
Puzzled by Fuji S2 Interpolation
« Reply #3 on: January 16, 2003, 08:46:33 am »

I'm puzzled by what someone told me about the Fuji S2 on another thread in this forum.  The S2 apparently interpolates all images up to 12 million "recording" pixels from the 6 megapixels that actually make up the sensor.  Therefore, 50% of all the information in the RAW or TIFF file is made up of imaginary pixels.  If the user wants a 6 megapixel file, it is interpolated down from 12 megapixels, i.e., 50% of the information is discarded.

What happens if the 50% of the pixels the camera discards are the original pixels?  Would you then be left with all imaginary pixels?  Or does the camera have some way of remembering which pixels are real, and which ones are imaginary (interpolated)?
Logged

Robert Roaldi

  • Guest
Puzzled by Fuji S2 Interpolation
« Reply #4 on: January 16, 2003, 04:55:14 pm »

It's the geometry. Fuji's CCD's pixels are laid out octogonally but all other devices, screens, printers etc. expect rectangular arrays of pixels, so there has to be a conversion. In their case, the camera does it. It must have been convenient for FUJI to interpolate to 12 meg by default for all images and then only res-down to 6 meg if you don't want/need the 12 meg.

It's a little like apples and oranges to compare what it does to other BAYER CCD's. Norman Koren himself states that the FUJI sensor a difficult thing to model for comparison purposes.

In the general sense, all digicams interpolate. That's how they get colours at each pixel location. Except for the Foveon, I suppose.

Another poster said it best. The final image is what counts. How the sensor delivers the end-result is interesting for its own technical self, of course. And it may help you to determine the limitations of your camera of choice.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Puzzled by Fuji S2 Interpolation
« Reply #5 on: January 19, 2003, 03:27:01 pm »

Quote
We see on the bottom of this page in Phil Askey's review that the S2 in 12 mp mode weighs in at 1800 horizontal and 1700 vertical, but only at 1650 and 1400 in 6 mp mode.
And these vertical and horizontal figures are better than I have seen for any existing 6MP "rectangular pixel array", while I think I have read that the resolution for 45 degree diagonals is less than for standard 6MP sensors. This all makes sense when you consider the pixel layout, which resembles that of the stars on the US flag: there are more vertical and horizontal lines (with less pixels along each line) but less diagonal lines than a "square" array of the same number of pixels.

I also remember reading that human vision resolves more detail with horizontal and vertical lines than with diagonals, so Fuji's arrangement might be well matched to its ultimate goal of images that look good to the viewer.

Yes, when constrained to use a rectangular array data format for output, more than 6MP are needed to avoid loss of real information: interpolation to twice as many pixels exactly fills in each gap between photosites on vertical and horizontal lines, so is the minimum that will avoid discarding any original photosite data. Put another way, "12MP square" is the minimum that has the same number of vertical and horizontal lines as "6MP octagonal".
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Puzzled by Fuji S2 Interpolation
« Reply #6 on: February 06, 2003, 04:23:02 pm »

Let me try to answer a bit the technical question, but I must first object to jargon like "real" and "imaginary" pixels for distinguishing between the values at photo-sites and the interpolated values at the "gaps" between them. ALL normal digital camera pixels are produced with some interpolation (since each records only one color), and with small enough photo-site size, they mostly give highly accurate information. Since even uninterpolated photo-site data are already at best only accurate, not exact, the "perfect/approximate" or "real/imaginary" dichotomy is an oversimplification. So I would describe pixels just as being of higher and lower quality according to the extent and quality of interpolation that has been applied.

My educated guess is that each of the three color channel values for each of the Fuji's 12 million output pixels is computed directly by interpolation from the raw values at several nearby photo-sites which records that color, rather than interpolating twice (first to fill in missing colors at each photo-site, then to fill in pixels in the gaps between the photo-sites.)

Given that, it is not clear whether the interpolated values at photosites are better or worse than the ones for the gaps. Probably, the best values are the ones at a photo-site for the color that it actually records  needs no interpolation, then the values at gaps are next best, and the values at photo-sites for the two colors not directly recorded there are actually the worst, since they have to be interpolated from a bit further away than in the case of interpolating into the gaps. So who can even say which pixels are best, let alone which are real and which are imaginary?

If your worry is that 12 million pixels are reduced to 6 million new, larger ones by just discarding half the pixels, relax, it doesn't work that way. Instead, the value in each color channel for each of the new, larger pixels is computed from a weighted average of values at all the nearby old, smaller pixels: weight greatest for the closest ones of course. So each of the final 6MP wil be computed from a mixture of values associated with photo-sites and values associated with the gaps.
Logged
Pages: [1]   Go Up