Ad
Ad
Ad
Pages: [1] 2 »   Bottom of Page
Print
Author Topic: Canon announces 50 MP prototype sensor  (Read 10109 times)
Digiteyesed
Full Member
***
Offline Offline

Posts: 159


WWW
« on: June 07, 2007, 09:21:42 AM »
ReplyReply

Gadget Lab:

"Just when you thought it was safe to ignore megapixels in favor of more meaningful measures of image quality, Canon goes and makes a 50 megapixel prototype to shunt jaws to floors. Claimed to be twice as sensitive as the nearest competition, the sample sensor is 19x28mm in size, the same dimensions already used in its fancy DSLR models.

There are no release plans announced, but when this arrives, it will result in pictures about 8,000 by 6,000 pixels in size: enough to make a billboard-sized poster at 20dpi, or a 4x6 at an amazing 1,333 dpi: does technology exist that can even print at such resolution at photo quality?"

---

Someone correct me if I'm wrong, but doesn't this photo site density greatly exceed the resolution of every 35mm lens currently on the market?
Logged

Neutral Hills Stills
A visual journey through this unique area of East Central Alberta, Canada.
AndyF2
Jr. Member
**
Offline Offline

Posts: 89


« Reply #1 on: June 07, 2007, 12:23:44 PM »
ReplyReply

Quote
Gadget Lab:

"Just when you thought it was safe to ignore megapixels in favor of more meaningful measures of image quality, Canon goes and makes a 50 megapixel prototype to shunt jaws to floors. Claimed to be twice as sensitive as the nearest competition, the sample sensor is 19x28mm in size, the same dimensions already used in its fancy DSLR models.

There are no release plans announced, but when this arrives, it will result in pictures about 8,000 by 6,000 pixels in size: enough to make a billboard-sized poster at 20dpi, or a 4x6 at an amazing 1,333 dpi: does technology exist that can even print at such resolution at photo quality?"

---

Someone correct me if I'm wrong, but doesn't this photo site density greatly exceed the resolution of every 35mm lens currently on the market?
[a href=\"index.php?act=findpost&pid=121605\"][{POST_SNAPBACK}][/a]
This may reduce the need for some long telephotos; a crop from the center of this sensor with a good lens, may be sharper cropped corner to cropped corner, than a full frame sensor corner to corner (assuming resolution near the center of a lens is significantly better than the corners).

However it's not the sensor I want  

How about a sensor where each pixel could be set with an exposure sensitivity before the image was taken?  

This allows an HDR image with a single exposure.  The raw file would contain the data for each pixel as it does now, and would also include each pixel's low/normal/high sensitivity setting.  Postprocessing can generate the HDR file from that information.  

It would not be necessary for the ISO of each pixel to be set to any value, and would instead have basic -4, 0 +4 exposure steps.  The camera would use a single-ISO sensor with ideally the same number of MP, or perhaps one quarter the MP, as it's full-frame exposure sensor.  The data from this is evaluated to find the blown and blocked areas of the image, then preset the corresponding pixels in the principal capture sensor to the required low/normal/high sensitivities.

If this amount of complexity and processing time seems excessive now, it will be feasible in a very small number of years.  
Andy Fraser
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #2 on: June 07, 2007, 11:15:36 PM »
ReplyReply

Quote
Someone correct me if I'm wrong, but doesn't this photo site density greatly exceed the resolution of every 35mm lens currently on the market?
[a href=\"index.php?act=findpost&pid=121605\"][{POST_SNAPBACK}][/a]

Resolution figures always have to be qualified by an MTF percentage. Quoted numbers of lines or 'line pairs per millimetre' by themselves are meaningless unless accompanied by some sort of contrast rating.

A 50mp sensor of dimensions 19x28mm should have a resolving power of approximately 126 lp/mm (at around 10% MTF) based on conservative evidence it usually takes about 2.5 pixels to record one line pair width (1.25 pixels per line).

I understand a lens that is diffraction limited at f8 can theoretically resolve about 97 lp/mm with a contrast loss of 50% (MTF 50%). At 30% MTF, resolution at a diffraction limited f8 would approach 140-150 lp/mm, at a guess.

I doubt whether any 35mm lenses are capable of true diffraction limited performance at f8, except perhaps in the dead centre of the image circle. But there probably are a few lenses which can resolve at f5.6 and f4 this theoretical diffraction limited resolution of f8.
Logged
500r420
Newbie
*
Offline Offline

Posts: 17


« Reply #3 on: June 12, 2007, 12:21:38 PM »
ReplyReply

Even if the pixel size of the sensor is smaller than the smallest point the lens can resolve, sensor pixels and points created by the lens do not line up perfectly, nor are they the same shape. Although the returns of denser sensors start to diminish as the pixel number starts getting very high, there will still benefits in total recorded detail even whilst the sensor is able to record many more lines than the lens
Logged
macgyver
Sr. Member
****
Offline Offline

Posts: 510


« Reply #4 on: June 12, 2007, 12:40:33 PM »
ReplyReply

Correct me if I am wrong but didn't they state that this prototype was for looking into enhanced survailence camera tech, not just standard DSLRs?
Logged
KAP
Full Member
***
Offline Offline

Posts: 168


WWW
« Reply #5 on: June 12, 2007, 02:45:32 PM »
ReplyReply

---

Someone correct me if I'm wrong, but doesn't this photo site density greatly exceed the resolution of every 35mm lens currently on the market?
[a href=\"index.php?act=findpost&pid=121605\"][{POST_SNAPBACK}][/a]
[/quote]

I don't think it matters as much as you might think. When scanning film, scanning at a higher resolution is better than scanning at a lower resolution then upsizing to make a big print, even if the lower res scan captures all the detail that's on the film, if you get what I mean. Or to put it another way, it's better having true fuzzy real information than adding some computer guess work.
IMHO.

Kevin.
Logged
feppe
Sr. Member
****
Offline Offline

Posts: 2909

Oh this shows up in here!


WWW
« Reply #6 on: June 12, 2007, 03:45:32 PM »
ReplyReply

Quote
I don't think it matters as much as you might think. When scanning film, scanning at a higher resolution is better than scanning at a lower resolution then upsizing to make a big print, even if the lower res scan captures all the detail that's on the film, if you get what I mean. Or to put it another way, it's better having true fuzzy real information than adding some computer guess work.
IMHO.

Kevin.

I'd be thrilled to hear some evidence to back this claim up. Yes, even anecdotal evidence counts as the claim is quite outrageous.
« Last Edit: June 12, 2007, 03:46:25 PM by feppe » Logged

500r420
Newbie
*
Offline Offline

Posts: 17


« Reply #7 on: June 12, 2007, 08:30:34 PM »
ReplyReply

Quote
I'd be thrilled to hear some evidence to back this claim up. Yes, even anecdotal evidence counts as the claim is quite outrageous.
[a href=\"index.php?act=findpost&pid=122479\"][{POST_SNAPBACK}][/a]

The explanation (i wouldn't say evidence) is in my post
he is correct

the point is that it doesn't capture all of the detail, even if the scanner has a line for each line on the negative
the "bottleneck" effect does not apply very well to lens/media as improving either will improve the overall recorded detail
Logged
dkeyes
Sr. Member
****
Offline Offline

Posts: 259


WWW
« Reply #8 on: June 13, 2007, 01:32:38 AM »
ReplyReply

Wouldn't it be better to have less pixel density on a larger sensor than have more pixel density on a smaller sensor? Obviously, the larger sensor would be coupled with larger format lenses as well.
For example, if lens and sensor quality were equal. Wouldn't a Mamiya zd back at 20mp (med. format) would be better than a Canon 1ds III at 20mp?
This scenario will most likely be tested in the near future.
- Doug
Logged

digitaldog
Sr. Member
****
Offline Offline

Posts: 9241



WWW
« Reply #9 on: June 13, 2007, 08:03:25 AM »
ReplyReply

Quote
How about a sensor where each pixel could be set with an exposure sensitivity before the image was taken? 

This allows an HDR image with a single exposure.

That would get my vote too. Enough pixels, that's kind of easy to make sales. Make the pixels BETTER! I'd prefer a 5D with true HDR over a huge sensor anyd day.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #10 on: June 13, 2007, 10:18:12 AM »
ReplyReply

Quote
How about a sensor where each pixel could be set with an exposure sensitivity before the image was taken? 

[a href=\"index.php?act=findpost&pid=121642\"][{POST_SNAPBACK}][/a]

How the heck could that work? Unless you are talking about the camera on a tripod where the sensor is able to preview the scene before the shutter is released, in which case the conditions are right for bracketed exposures and image blending. It would be disastrous if individual pixels were to auto-adjust their sensitivity for a particular scene, then movement of the subject or target were to change the conditions. Even very minor movement could cause insensitive pixels to be aligned with low level signals and highly sensitive pixels aligned with specral highlights.

I suppose we could invoke some of the more colorful theories that attempt to explain quantum weirdness whereby all light signals send out an exploratory signal first to check out the path the real light wave is going to embark upon, then return in a reversal of time to inform the signal proper what's in store for it (just in case some physicist is attempting the two slit experiment).
Logged
500r420
Newbie
*
Offline Offline

Posts: 17


« Reply #11 on: June 13, 2007, 10:39:21 AM »
ReplyReply

Quote
How the heck could that work? Unless you are talking about the camera on a tripod where the sensor is able to preview the scene before the shutter is released, in which case the conditions are right for bracketed exposures and image blending. It would be disastrous if individual pixels were to auto-adjust their sensitivity for a particular scene, then movement of the subject or target were to change the conditions. Even very minor movement could cause insensitive pixels to be aligned with low level signals and highly sensitive pixels aligned with specral highlights.

I suppose we could invoke some of the more colorful theories that attempt to explain quantum weirdness whereby all light signals send out an exploratory signal first to check out the path the real light wave is going to embark upon, then return in a reversal of time to inform the signal proper what's in store for it (just in case some physicist is attempting the two slit experiment).
[a href=\"index.php?act=findpost&pid=122611\"][{POST_SNAPBACK}][/a]

the camera can preview the scene via a sophisticated light meter
or you could set regional sensitivity to sorta have a split filter without having one
Logged
EricV
Full Member
***
Offline Offline

Posts: 136


« Reply #12 on: June 13, 2007, 12:03:15 PM »
ReplyReply

Quote
How about a sensor where each pixel could be set with an exposure sensitivity before the image was taken? 

This allows an HDR image with a single exposure.  The raw file would contain the data for each pixel as it does now, and would also include each pixel's low/normal/high sensitivity setting.  Postprocessing can generate the HDR file from that information. 

Andy Fraser
[a href=\"index.php?act=findpost&pid=121642\"][{POST_SNAPBACK}][/a]

There are already sensors which have considerably greater dynamic range than those currently used in consumer digital cameras.  There are several methods of providing this increased range:

1) Large pixels with high full-well capacity, low noise electronics, digitize 16 bits.
2) Split pixels, each pixel has large and small part, or high and low gain readout.
3) Electron-multiplying CCD, greatly reduced noise, digitize 16 bits.
4) Pixels with logarithmic rather than linear response, digitize 12 bits.

Some cameras based on these sensors are already produced for specialized markets (scientific research, low-light surveillance), but so far most lack the high pixel counts needed for general photography.
Logged
KAP
Full Member
***
Offline Offline

Posts: 168


WWW
« Reply #13 on: June 13, 2007, 01:11:16 PM »
ReplyReply

Quote
I'd be thrilled to hear some evidence to back this claim up. Yes, even anecdotal evidence counts as the claim is quite outrageous.
[a href=\"index.php?act=findpost&pid=122479\"][{POST_SNAPBACK}][/a]

Well I'm making some 350 meg scans (actualy 700meg 48 bit) right now, even at 100% they hold up well, 1DsmkII up-sized with software don't come close. Even scanning a section at 8000dpi looks better than it has a right to, any softness is created by the taking lens and just gets enlarged.
 If it was software upsized you can add a heap of guessed at pixels as well. The thing with the lens softness is if you can put up with it at A3 you can put up with it at 50inches, it still looks like a real image. Building's edges are still sharp, there's not the smoothing out software creates. Software up-sizing depends a lot on the subject as to how successful it is. I've on occaison been dissapointed with 30 inch prints from the Canon, knowing a good 6x7 would of been much better. Then again some Canon files can just keep going bigger.
I think the same will be with more pixels, the lens will not be any better or worse than they are now, it will look just as good at A0 as it does now at A3, I can't think of many commercial situation where we get the max lpm from our lenses I certainly don't as I'm shooting from helicopters. A good 6x7 will enlarge much much better than a 1DsmkII. Plus more pixels done well should mean smaller artfacts etc and that will mean better up-sizing with software should you require.

Kevin.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #14 on: June 13, 2007, 07:47:28 PM »
ReplyReply

Quote
the camera can preview the scene via a sophisticated light meter

[a href=\"index.php?act=findpost&pid=122617\"][{POST_SNAPBACK}][/a]

Yes, of course it can. However, if the sensitivity of individual pixels is going to match the intensity of the light hitting them, you need the camera to be on a tripod and the subject to be still, otherwise you could get worse dynamic range and lots of noise in parts of the image that did not correspond exactly with the preview, not to mention wrong colors, halos and effects much worse than color fringeing.
Logged
500r420
Newbie
*
Offline Offline

Posts: 17


« Reply #15 on: June 13, 2007, 07:52:37 PM »
ReplyReply

Quote
Yes, of course it can. However, if the sensitivity of individual pixels is going to match the intensity of the light hitting them, you need the camera to be on a tripod and the subject to be still, otherwise you could get worse dynamic range and lots of noise in parts of the image that did not correspond exactly with the preview, not to mention wrong colors, halos and effects much worse than color fringeing.
[a href=\"index.php?act=findpost&pid=122709\"][{POST_SNAPBACK}][/a]

not if the meter gave the information to the sensor just before the picture was taken
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #16 on: June 13, 2007, 08:37:19 PM »
ReplyReply

Quote
Well I'm making some 350 meg scans (actualy 700meg 48 bit) right now, even at 100% they hold up well, 1DsmkII up-sized with software don't come close. Even scanning a section at 8000dpi looks better than it has a right to, any softness is created by the taking lens and just gets enlarged.
 [a href=\"index.php?act=findpost&pid=122649\"][{POST_SNAPBACK}][/a]

I can believe this is generally true. I have a number of 35mm slides and negatives which I tend to rescan every time I get a higher resolving scanner.

The first scans were done by Kodak with their PhotoCD system at 2000 ppi. I later bought my first scanner, the Nikon LS 2000 which scans at 2700 ppi and noticed marginally sharper results allowing me to make larger, better looking prints.

My next scanner had an optical resolution of 4000 ppi and was able to deliver even more detail, or at least better defined detail from those slides and negatives that were particularly sharp.

My current scanner is the Dimage Elite 5400 II which, at 5400 ppi, produces the best result of all. I have no hesitation in making 22"x33" prints from the sharper of the negatives and slides in my archives.

This is what I believe is happening and once again the explanation lies in the MTF response of the scanner lens.

A scanner resolution of 2000 ppi equates to 40 lp/mm; 2700 ppi equates to 54 lp/mm; 4000 ppi to 80 lp/mm and 5400 ppi to 108 lp/mm.

The problem is that no lens can deliver such resolution at 100% MTF, so in practice the resolution limits are significantly lower. I'd say they are more like 30 lp/mm, 40 lp/mm, 60 lp/mm and 75 lp/mm.

It's possible to record 75 lp/mm on 35mm film, especially if the film is fine grained like T-Max 100, the target contrasty and the shutter speed fast enough etc., but it's probably not possible to capture such resolution in a scan of the film unless the scanner is theoretically capable of a much greater resolution than 75 lp/mm.
Logged
AndyF2
Jr. Member
**
Offline Offline

Posts: 89


« Reply #17 on: June 13, 2007, 10:08:52 PM »
ReplyReply

My original thought was the exposure evaluator would be a second sensor with the same resolution as the image sensor, but this requires both be aligned within one pixel tolerance.  Obviously too difficult.  There is also the problem Ray brought up, with the scene changing slightly between measurement and exposure.  Bright objects might move into pixel regions that were expected to be dark.

So the image sensor itself would need to include the sensitivity measurement.  During a preliminary exposure each pixel would discover whether it will saturate and if so, configure itself for a reduced sensitivity.  One method is to incorporate a mask on the surface of the pixel.  Mask 75% at that site, and that's a 2 stop sensitivity drop.  The remaining 25% of the pixel still collects the pixel's worth of resolution available at that site.  The mask would be something similar to LCD technology, transparent or blocking.  Pixels then have -0 or -2 stop sensitivities.

The evaluation still has to be a seperate exposure but the evaluation and image exposures could be done within about 3x the shutter time selected.  This shortens the time gap during which the boundaries of bright and dark objects can move, and result in a blown pixel if a cell which had darkened itself, ended up being exposed beyond it's limit.  Still an interesting new exposure failure mode!

It seems there will always be challenges with HDR unless the immge can be taken with a single shot, by either pixels scaling themselves during the exposure, or cells having a large dynamic range and low noise to begin with.  50 MP is impressive but on closer thught, technically a simpler achievement.  What would be truly desirable is a slightly lesser resolution, 14-16 bit range, very low noise (1 or 2 bits?), and (not mentioned yet) nearly identical sensitivity across all pixels of a given sensor.

Andy
Logged
BJL
Sr. Member
****
Offline Offline

Posts: 5171


« Reply #18 on: June 25, 2007, 03:38:00 PM »
ReplyReply

Quote
... 50 megapixel ... 19x28mm in size

Someone correct me if I'm wrong, but doesn't this photo site density greatly exceed the resolution of every 35mm lens currently on the market?
[a href=\"index.php?act=findpost&pid=121605\"][{POST_SNAPBACK}][/a]
That sensor is apparently intended for surveillance cameras, where a custom lens of fixed focal length and a single optimal aperture can probably be used. I would guess that such a lens can probably match the resolution of the roughly 3.5 micron pixel spacing of this sensor. (Maybe good Canon primes can too: about 200 to 300 lp/mm is needed.)

As an aside, Canon's use yet again of 19x28mm "1D" format seems to confirm that this is the largest sensor that Canon can make with special more expensive multiple exposure methods: no increase in that size limit in the three years since teh 1DMkII sensor.
Logged
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #19 on: June 25, 2007, 08:43:17 PM »
ReplyReply

Quote
This may reduce the need for some long telephotos; a crop from the center of this sensor with a good lens, may be sharper cropped corner to cropped corner, than a full frame sensor corner to corner (assuming resolution near the center of a lens is significantly better than the corners).[a href=\"index.php?act=findpost&pid=121642\"][{POST_SNAPBACK}][/a]

What you get is basically the performance of a smaller-sensor camera when you crop, and wide lenses for FF format tend not to be very sharp, so you're really better off with the long lens and using the full frame.
Logged
Pages: [1] 2 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad