Sharp's new sensor - just curious

Don Fraser · Nov 8, 2006

I note that in the news item today, Sharp's new sensor will have a microsite of 1.88 microns.

I was wondering how this compares to a typical average grain size on a black and white film, say Tri-X, developed normally. Are the two sizes close, or is the B + W film grain size of a different order of magnitude?

I'm not trying to start another image quality comparison. Just curious about the photosite/grain size comparison.

Cideway · Nov 8, 2006

Well if you cook your TriX and push it to 3200 you may get close to the amount of grain in that this sensor may produce... at 400 ISO
--
Not long now till i get a special K ... 10D

Info thread for new Pentax SLR users:
http://forums.dpreview.com/forums/read.asp?forum=1036&message=19455961

GMT +9.5

john.p.jones · Nov 8, 2006

Almost everyone here agrees that pushing these small sized sensors to higher resolutions is foolish.

What I was wondering is whether the techniques manufacturers are using to achieve 12MP on these tiny sensors could be applied to produce better 6-7MP sensors then were produced when that was the edge? Are the fruits of all this research retargetable to a better goal (better performance without more pixels) or are they just wasted effort only useful for producing noisy high density sensors?

Roland Karlsson · Nov 8, 2006

Don Fraser said:
I note that in the news item today, Sharp's new sensor will have a
microsite of 1.88 microns.

I was wondering how this compares to a typical average grain size
on a black and white film, say Tri-X, developed normally. Are the
two sizes close, or is the B + W film grain size of a different
order of magnitude?

I'm not trying to start another image quality comparison. Just
curious about the photosite/grain size comparison.

The comparison does not hold with current technology.

Larger pixels means less noise overall, i.e. small pixel cameras is noisier.

Larger grains means more noise overall, i.e. small grain films is cleaner.

SciFi:

If each cell was a photon counter (with zero readout noise) and the fill factor could be kept near to 100% then you shoud have the same overall noise independent of pixel size.

--
Roland

BJL · Nov 8, 2006

Roland Karlsson said:
SciFi:

If each cell was a photon counter (with zero readout noise) and the fill factor could be kept near to 100% then you shoud have the same overall noise independent of pixel size.

Yes, if one could eliminate read-out noise, extremely tiny photosites each detecting one photon would be a lot like film grains. Film grains need to be hit by about five photons to become "exposed".

And even though each individual silver halide grain ("chemical pixel") has very poor dynamic range and S/N ratio (about 2:1?), so that viewing at 100% pixels shows a sea of noise, printing those very high "pixel" count images at many thousands of "pixels" per inch can smooth out noise nicely, and give wide dynamic range. This is a kind of dithering.

Likewise, with negligible read-out noise, even tiny pixels in very large quantities printed at high PPI could give good dynamic range and noise levels. We must avoid the mistake of judging images with more, smaller pixels at equal PPI, which is like comparing different films by viewing prints of different size and degree of enlargement.

Roland Karlsson · Nov 8, 2006

BJL said:
...

You can take it even one step further. Lets assume that the sensor has no pixels at all. Every photon that hits the surface makes an impression exactly where it hits and the impression is proportional to the energy. To read the picture you use some kind of technique to localize all the hits and their amplitude.

In this case you will have all information - except phase and direction for the photons. And of course - to save position and amplitude for zillions of photons you will need weeks to download the picture and rooms full of storage

--
Roland

GURL · Nov 10, 2006

Below is a lateral scan (slice) through an emulsion layer of developed APX100 film.

(from Black and White Film emulsions: the state of the art - A report by Erwin Puts - http://www.imx.nl/photosite/technical/Filmbasics/filmbasics.html )

Grain: the individual silver halide crystals. The average size of such a crystal is 0,2 to 2 micron.

Each grain of silver halide develops in an all-or-nothing way.

The smaller the crystals, the finer the detail in the photo and the slower the film.

As the film becomes progressively more exposed, each incident photon is less likely to impact a still-unexposed grain, yielding the logarithmic exposure behavior.

At 4000 samples per inch a Kodachrome scan produces roughly 21 megapixels from a 35mm frame, outperforming all current 35mm DSLRs. Going even further, professional scanners capable of 8000 or 12,000 spi turn a Kodachrome's native resolution into a sharp 85 to 192 megapixel file.
--
Georges Lagarde http://www.panorama-numerique.com

Roland Karlsson · Nov 10, 2006

GURL said:
At 4000 samples per inch a Kodachrome scan produces roughly 21
megapixels from a 35mm frame, outperforming all current 35mm DSLRs.
Going even further, professional scanners capable of 8000 or 12,000
spi turn a Kodachrome's native resolution into a sharp 85 to 192
megapixel file.

This has been proven again and again and again to be untrue. Both a 20 Mpixel scan and absolutely a 100 MPixel scan of color film yields very grainy and unsharp results.

--
Roland

GURL · Nov 10, 2006

Roland Karlsson said:
This has been proven again and again and again to be untrue. Both a
20 Mpixel scan and absolutely a 100 MPixel scan of color film
yields very grainy and unsharp results.

A single pixel is worth many silver halide grains: a grain is kept or removed, present or absent, 1 or 0, while a single photosite will return one among many possible values (the interval is 0..255 or even larger.)

In a fews words, comparing grain size and photosite size is not enough...
--
Georges Lagarde http://www.panorama-numerique.com

BJL · Nov 10, 2006

GURL said:
Below is a lateral scan (slice) through an emulsion layer of developed APX100 film.

Which shows why film images should never be displayed at "100% pixels": noise worse that a cheap nasty little digicam is ISO 3200! The ultimate resolution limits of silver-halid sensors are very high, but at the cost of miserably low S/N ratio and dynamic range if you view large enough for the individual pixels to be anywhere closer visible.

However, that does to really matter in practice: normal printing is at very high "PPI", so that the eye's blurring together of numerous nearby "pixels" greatly improves the noise levels and dynamic range. I call this "dithering", though that might not be correct technical usage.

A similar thing is probably true for electronic sensors with very many, very small photo-sites. In each case, it is very misleading to look at the ultimate resolution limit set by grain size or photo-site size, or at the noise levels seen when one enlarges to the "visible pixelation" limit.

GURL said:
Grain: the individual silver halide crystals. The average size of such a crystal is 0,2 to 2 micron.

Interesting: electronic sensors are into that range now (with far better sensitivity and S/N ratio).

John Sheehy · Nov 10, 2006

BJL said:
Likewise, with negligible read-out noise, even tiny pixels in very
large quantities printed at high PPI could give good dynamic range
and noise levels. We must avoid the mistake of judging images with
more, smaller pixels at equal PPI, which is like comparing
different films by viewing prints of different size and degree of
enlargement.

True. The problem with most web-based proofs and comparisons is the discrete nature of monitor pixels. If you took a camera that kept the same lens, but doubled the MP in the next version, how do you properly compare them? If you shrink images from both to the same number of pixels as a web image, the resolution benefits of the higher-MP camera are all lost. If you compare 100% crops, there will be less pixel-to-pixel contrast in the higher-MP one, and more noise intensity at the same exposure level. The only way to properly compare is to make large prints with the same subject magnification, and view from a distance, or upsize both for monitor display so that neither benefits the psychological benefit of having any detail near the screen mode's maximum resolution, and stand 20 feet away from the monitor.

If we had monitors with extremely high resolution, where you could have two different PPIs on the screen at once, we could really see the benefits or disadvantages of pixel size trade-offs. Our common medium of demonstration is totally inept to show the differences in a traditional manner.

--
John

John Sheehy · Nov 10, 2006

I said:
True. The problem with most web-based proofs and comparisons is
the discrete nature of monitor pixels.

... or screen mode pixels.

--
John

Andrew dB · Nov 10, 2006

BJL said:
GURL said:

Below is a lateral scan (slice) through an emulsion layer of developed APX100 film.

Click to expand...

Which shows why film images should never be displayed at "100%
pixels": noise worse that a cheap nasty little digicam is ISO 3200!
The ultimate resolution limits of silver-halid sensors are very
high, but at the cost of miserably low S/N ratio and dynamic range
if you view large enough for the individual pixels to be anywhere
closer visible.

However, that does to really matter in practice: normal printing is
at very high "PPI", so that the eye's blurring together of numerous
nearby "pixels" greatly improves the noise levels and dynamic
range. I call this "dithering", though that might not be correct
technical usage.

A similar thing is probably true for electronic sensors with very
many, very small photo-sites. In each case, it is very misleading
to look at the ultimate resolution limit set by grain size or
photo-site size, or at the noise levels seen when one enlarges to
the "visible pixelation" limit.

BJL said:

Grain: the individual silver halide crystals. The average size of such a crystal is 0,2 to 2 micron.

Click to expand...

Interesting: electronic sensors are into that range now (with far
better sensitivity and S/N ratio).

Specialised emulsions can apparently be made with grain sizes of the order of a few nanometers but the loss of film speed is massive and outside of studio still life, they would be difficult to use in general photography.

John Sheehy · Nov 10, 2006

Roland Karlsson said:
BJL said:

...

Click to expand...

You can take it even one step further. Lets assume that the sensor
has no pixels at all. Every photon that hits the surface makes an
impression exactly where it hits and the impression is proportional
to the energy. To read the picture you use some kind of technique
to localize all the hits and their amplitude.

I said something like this in open talk a week ago, but I didn't get much response.

Roland Karlsson said:
In this case you will have all information - except phase and
direction for the photons. And of course - to save position and
amplitude for zillions of photons you will need weeks to download
the picture and rooms full of storage

If it were a low-light exposure, though, the data set would be rather small.

--
John

D Q E · Nov 10, 2006

Cideway said:
Well if you cook your TriX and push it to 3200 you may get close to
the amount of grain in that this sensor may produce... at 400 ISO

And this bold statement is based on what evidence, precisely? I reckon it's a dead cert that a piece of Tri-X at 7.1mm x 5.3mm would produce abysmal pictures of virtually any subject. You can see just how bad by looking at some Tri-X Super 8 and then cropping a little bit.

--
mumbo jumbo

BJL · Nov 10, 2006

John Sheehy said:
the discrete nature of monitor pixels. If you took a camera that kept the same lens, but doubled the MP in the next version, how do you properly compare them? If you shrink images from both to the same number of pixels as a web image, the resolution benefits of the higher-MP camera are all lost.

Roughly, but maybe not exactly when dealing with standard Bayer CFA pixels and the loss of detail in "demosaicing". A 6MP Bayer CFA sensor probably only resolves about as much as about 3 million full three color pixels (like those on a monitor or a Foveon style "X3" sensor), and it seems that 6 million "single color pixels" processed to produce 3 million "three color pixels" can retain that resolution of 3 million full three color pixels, such as one might get with a 3MP X3 Foveon sensor or a 3MP film scan.

Some examples

the 13.5MP Kodak 14/N output at 6MP had noticeably better resolution than 6MP DSLRs.
The 5MP output mode of Sony's former digicam with 8MP 2/3" CCD showed more resolution than 5MP images from Sony's previous 5MP 2/3" CCD digicam. Then again, those images also showed slightly more noise even when equalized at 5MP, so the noise/detail trade-of did not noticeably change.

DRG · Nov 10, 2006

Roland Karlsson said:
In this case you will have all information - except phase and
direction for the photons.

Phase and photon direction aren't too separate concepts. Either you think of photons in terms of count and direction or you think of the EM wave which has amplitude and phase. In this latter point of view, "rays" of photons converging on a point on a sensor from different directions merely looks like an incident plane wave but with the wavefront curved in a semicircle with its center at the point of convergence.

If we could capture phase, then we could forget about focus altogether. All the optical information about the subject would be available and focus could recreated for any plane after the fact. In fact, we'd have a full-color hologram.

David

D Q E · Nov 10, 2006

GURL said:
Roland Karlsson said:

This has been proven again and again and again to be untrue. Both a
20 Mpixel scan and absolutely a 100 MPixel scan of color film
yields very grainy and unsharp results.

Click to expand...

A single pixel is worth many silver halide grains: a grain is kept
or removed, present or absent, 1 or 0, while a single photosite
will return one among many possible values (the interval is 0..255
or even larger.)

In a fews words, comparing grain size and photosite size is not
enough...

Well said, this fact is ALWAYS ignored by the 'resolution' crowd.

--
mumbo jumbo

Eric Fossum · Nov 11, 2006

I am really going to have to get my gigapixel digital film sensor paper posted soon. The paper proposes using binary pixels (e.g. geiger-mode photodiodes) that I call jots and that are very very small. The jots are readout and OR'd together to create a grain, and the grains are then digitally developed by software to create pixels and an image. Grain size can be dynamically varied, both inter-frame and intra-frame.

This paper was presented at the 2005 IEEE Workshop on CCDs and Advanced Image Sensors in Karuizawa, Japan and also at the Shizuoka University symposium on Nanoelectronics in Nov 2005 (I think).

I will try to put it up on the Siimpel site. Siimpel owns the pending patent.

-Eric

Joe0Bloggs · Nov 11, 2006

http://www.sciencedirect.com/science?_ob=ArticleURL&_udi=B6TVD-41MJ73V-20&_coverDate=02%2F28%2F1998&_alid=486110289&_rdoc=1&_fmt=&_orig=search&_qd=1&_cdi=5532&_sort=d&view=c&_acct=C000050221&_version=1&_urlVersion=0&_userid=10&md5=ed7f7be5f495c7b83c8a19fc54a56b98

"The novel type of the Silicon Photodiode — Limited Geiger-mode Photodiode (LGP) has been produced and studied. The device consists of many ≈104 mm−2 independent cells ≈10 mkm size around n+ -“pins” located between p-substrate and thin SiC layer. Very high gain more than 104 for 0.67 mkm wave length light source and up to 6·105 for single electron have been achieved. The LGP photon detection efficiency at the level of

Eric Fossum said:
one percent

has been measured."

I can see why this kind of sensor hasn't gone to market...

Sharp's new sensor - just curious

Senior Member

Veteran Member

Forum Enthusiast

Forum Pro

Veteran Member

Forum Pro

Leading Member

Forum Pro

Leading Member

Veteran Member

Forum Pro

Forum Pro

Leading Member

Forum Pro

Senior Member

Veteran Member

Veteran Member

Senior Member

Senior Member

Veteran Member

Similar threads

Keyboard shortcuts