Well, your argument is sound. They are interpolating data between points, and this method does cause sharpness artifacts, but what I'd like to point out is that every camer interpolates between points. Let me give you a sample problem: you have a 2x * 2y matrix of sensors where each sensor senses an amount of light filtered by a color (typically in a CFA). Say that you want to produce a 2x * 2y matrix of pixels that are in full 3 channel, 24-bit color. Where do the other color channels come from? Each sensor only measures one color axis. The answer is that the other colors come from interpolating with the neighboring sensor values.
Right in that it is interpolated data. Wrong in that you can do the interpolations in any direction. Remember that your output must be in the form of a rectangular grid, diagonally interpolating would end up with a grid much like Fuji's honeycomb pattern. In this case, however, your pixels would not fit on a rectangular grid. Fuji is simply changing the number of unique sensors that need to be used for a given pixel.
ixel ratio) unlike CCDs with a square pattern, where the regions measured are offset by an actual pixel and have worse color aliasing. Color aliasing is possible to fix with crafty software or hand touchups.Actually, 2X digital zoom is very different from what Fuji is doing. Fuji generates pixels on a square grid from a diagonal grid which allows more pixels per source gemetrically, whereas 2X digital zoom actually uses half of the sensors. Scaling up an image by 2X afterward would be a better analogy, but it wouldn't be as smooth as having values on a diagonal grid to interpolate from, which is where the genius of the Fuji engineers came in. The fact that Phil compares the Fuji S1 Pro to the Nikon D1X is testiment to the fact that Fuji's engineers did a pretty good job, although clearly the D1X has more resolution, especially because it has more sensors. Still, Fuji's sensor pattern does allow for larger images to be created with less interpolation than the competition, leading to a higher quality image than some of the same resolution cameras (where resolution is in the number of sensors, and not the number of output pixels).
Which is what Phil holds them accountable for in his reviews... We hold them to the same standards, too.
I would prefer to take photographs with the higher resolution in camera, and use software to downsample afterwards. One can beat out JPEG artifacts like that. Assuming there is no RAW format... I would personally use a RAW format as you don't lose any data, and only have to store the uninterpolated sensor values.
Well, your argument is sound. They are interpolating data between
points, and this method does cause sharpness artifacts, but what
I'd like to point out is that every camer interpolates between
points. Let me give you a sample problem: you have a 2x * 2y matrix
of sensors where each sensor senses an amount of light filtered by
a color (typically in a CFA). Say that you want to produce a 2x *
2y matrix of pixels that are in full 3 channel, 24-bit color. Where
do the other color channels come from? Each sensor only measures
one color axis. The answer is that the other colors come from
interpolating with the neighboring sensor values.
Right in that it is interpolated data. Wrong in that you can do the
interpolations in any direction. Remember that your output must be
in the form of a rectangular grid, diagonally interpolating would
end up with a grid much like Fuji's honeycomb pattern. In this
case, however, your pixels would not fit on a rectangular grid.
Fuji is simply changing the number of unique sensors that need to
be used for a given pixel.
Fuji changed their orientation to along the diagonal so they could
interpolate more pixels legitimately from their sensors. It's quite
clever, and you don't lose much. It's still prone to errors, but I
give them credit for ensuring that the regions overlap along a
square border at a 1:1 (sensor
square pattern, where the regions measured are offset by an actual
pixel and have worse color aliasing. Color aliasing is possible to
fix with crafty software or hand touchups.
.......
Actually, 2X digital zoom is very different from what Fuji is
doing. Fuji generates pixels on a square grid from a diagonal grid
which allows more pixels per source gemetrically, whereas 2X
digital zoom actually uses half of the sensors. Scaling up an image
by 2X afterward would be a better analogy, but it wouldn't be as
smooth as having values on a diagonal grid to interpolate from,
which is where the genius of the Fuji engineers came in. The fact
that Phil compares the Fuji S1 Pro to the Nikon D1X is testiment to
the fact that Fuji's engineers did a pretty good job, although
clearly the D1X has more resolution, especially because it has more
sensors. Still, Fuji's sensor pattern does allow for larger images
to be created with less interpolation than the competition, leading
to a higher quality image than some of the same resolution cameras
(where resolution is in the number of sensors, and not the number
of output pixels).
Which is what Phil holds them accountable for in his reviews... We
hold them to the same standards, too.
I would prefer to take photographs with the higher resolution in
camera, and use software to downsample afterwards. One can beat out
JPEG artifacts like that. Assuming there is no RAW format... I
would personally use a RAW format as you don't lose any data, and
only have to store the uninterpolated sensor values.
-Mike
Still, it reminds me of the audiophile with his 10-100k pancake
flat amp pushing into 20-40k not so flat speakers, listening with
30-20k ears in a room with response curves like a rollercoaster.
All in all I'd say todays average 3-4MP cams produce data that is
beyond the capacity of the average printer or video displays
capability, and beyond the capacity of the eye (other than nose to
paper w/loupe) to comprehend.
Does the music sound good? Do the prints look good? Do those you
share you music or images reward you with ohhhhh and ahhhhhh?
If yes.. then it is mission accomplished!, regardless of the
technical perfection that the finsihed product started from.
My 2.02352345769568534097 cents worth
Max
Max
Actually Geo, it's not surprising at all that you get excellent images with the Pro-70. Back in the "stone-age" of digital photography, (your know, about 3 years ago
) the Pro-70 and the Sony DSC-D700 were two of the finest digicams available for under about $25,000. The Pro-70 was unique in a way in that Canon saw fit to include in their firmware distortion correction so that 28mm images shot with the Pro-70 were exceptionally clean and it became the "darling" of the real-estate photographers who could just shoot and print and forget mousing around in the digital darkroom as we who had chosen the Sony competition had to do to get the same or similar quality.
Thanks, I try.
Sharpening artifacts generally come from using too much sharpening. There are (generally speaking) more sharpening artifacts when you use more interpolation, which is why artifacts become a problem when using the normal or strong modes as you mention above. Fuji has a smaller ratio of sensors to pixels when in its 6MP mode. Sharpening these edges when you're interpolating colors can yield poor color aberrations.
I'm not saying that it's not interpolation. It's just not the same kind of interpolation. This interpolation happens from a different layout of sensors. The output resolution of the camera is natively 6MP in the rectangular format that we are used to. The fact that it was only measured with 3.3 Million sensors means that that is the fundamental function limiting its fidelity, but the 6MP image is definately less interpolated than a competing camera whose once-interpolated 3.3MP image is pushed up to 6MP. That competing camera has a grid of 3.3MP that it must then interpolate between to get to 6MP. Fuji interpolates directly to 6MP. Each pixel of the 6MP output corresponds physically to an overlapped area of CCD sensors. Whereas on a Bayer-pattern square CCD, the same is true for the same resolution (3.3 Million Sensors maps to 3.3 Million Pixels).
Not quite losing information, just resizing down from interpolated up data. You interpolate up to fit it to a rectangular grid, then you resample that higher resolution grid down to 3.3MP. Remember that the 6 Million Pixels were created from 3 Million sensors. You didn't actually have that much information, you are just reinterpreting the information. Think about it like interpolating 3.3Million Sensors to 3.3Million Pixels like the rest of the cameras, only these sensors overlapped a little better.
--Geo Paris/FR
est superbe (especially in original size, fastly downloaded with a
broadband connecton)! I am amazed how good can be some pictures
taken with a 1.5 mp. Some of my best pictures printed in 8x10 have
been taken with my Olympus D600 (1.4 mp), some of these days I will
publish tham also on pbase.
I have also noticed that good B&W pictures can be apparently more
easily obtained from low resolution digital camera than from higher
resoluton ones. I guess that to palliate the low resolution, the
manufacturers have made a particular effort (lens + CCD electronic)
to get more apparent acutance/contrast hiding the low resolution
and this provides directly more vivid and impressive pictures at
small print size (up to 8x10) and in B&W.
This is something existing also in film camera. The pictures taken
by a medium format cameras seem often softer, if applied the same
processing, than those taken by 35 mm cameras. And the best,
sharpest, from far, 35 mm lenses, the Leica lenses, are not rated
with a high optical resolution in lab. tests...
This being said, with a medium format film camera, by pushing a
little the contrast in the dark room, you can get the level of
sharpness you want, have more dynamic range, print much larger
size, etc. It is a comfort that according to the kind of pictures
you do is just a comfort or is necessary for your goal. For
instance, the man who has been may be the greatest photographer
ever, our fellow French citizen Henri Cartier-Bresson, used only a
rangefinder Leica with always the same 50 mm lens, and never
cropped his pictures. But, to get the incredible pictures of Ansel
Adams, a large format camera or at least a medium format camera was
needed (in addition to genius!).
Same thing for digital... One can take great pictures with a Pro
70, yours prove that, other will need at least a 20 mp digital back
on their Hasselblad.
Jean-Paul
*
--Geo Paris/FR
--Geo Paris/FRActually Geo, it's not surprising at all that you get excellent
images with the Pro-70. Back in the "stone-age" of digital
photography, (your know, about 3 years ago
Sony DSC-D700 were two of the finest digicams available for under
about $25,000. The Pro-70 was unique in a way in that Canon saw fit
to include in their firmware distortion correction so that 28mm
images shot with the Pro-70 were exceptionally clean and it became
the "darling" of the real-estate photographers who could just shoot
and print and forget mousing around in the digital darkroom as we
who had chosen the Sony competition had to do to get the same or
similar quality.
I still get incredibly good results with my Sony (the D700 was
replaced with the D770) 1.5 megapixel camera. In fact, I recently
received an "Editor's Choice" award on Nature Photographer's Online
Magazine for an image I shot with my D770 (shown below). This
image, by the way, prints beautifully at 8x10. The "secret" to
getting good larger prints with a low resolution instrument is to
realize that one must "concentrate" those scant pixels so that they
are optimally used on a smaller target (as you did in your very
excellent candid portrait). You certainly wouldn't want to
photograph a basketball team and attempt to enlarge, but when 1.5
million pixels are concentrated in a reasonably small area, the
results can be very nice indeed.
Lin
--
http://204.42.233.244
