I love the performance of these larger sensors, and these demos blow everything I've seen out of the water, but I'd never carry that kind of gear around for any of my purposes (the size and weight more than the price, though that's definitely a large factor).
Have you ever played with a Sony RX-100? Nowhere near full frame, but like four times the sensor area of a typical pocket camera. I got one recently and it's nothing short of astonishing for what it looks like on the outside and how it feels to use. Even the AVCHD video capabilities are damn impressive (again, for its price and form factor).
Moonlight is for the most part just reflected sunlight, so it's not surprising that the colors are very similar. Obviously it's much less bright, by a factor of 500k according to Wikipedia, but that's what this sensor is about.
You can get similarly startling images from a plain point and shoot camera by installing it on a tripod and taking long exposure shots of things. You need 30 to 60 seconds exposure, sometimes available as a "starry sky" scene mode. Architecture is particularly suited as it tends not to move around. :) Even a cheap, small, P&S sensor is much more sensitive to light when you expose for 60 seconds, all while being completely noise-free.
The sensitivity didn't increase by exposing it longer - its simply accumulating more signal over time.
Many sensors are ill-suited to long exposures though. While they can do it, you end up with heat related issues (commonly manifested as stuck pixels and general noise). CCDs also end up saturating and spilling over to adjacent cells, making high dynamic scenes a challenge.
In a way it's kind of simple, isn't it? Having "only" Full HD on a 35 mm sensor simply means you have much larger individual potential wells, capturing more photons per pixel.
What I'm really wondering about is the dynamic range - wouldn't that also drastically increase?
Lenses are very constrained by physics, and cost. Ground glass and more exotic materials are expensive to produce - injection molded plastic isn't. A high MP sensor is miles cheaper than an upgraded optical system.
Higher pixel counts give less noise assuming the total light collection area is the same. Normally you get a bit less light collection area due to the gaps between pixels with a higher resolution sensor but overall I think higher pixel count generally wins, especially when working at the low iso end. So as long as the cost is reasonable why not stick lots of pixels in.
No, smaller pixels have more noise (on a pixel basis) because the incoming photons are averaged over a smaller area.
Of course, if you have more resolution, you can reduce the resolution in post-processing, and that will also reduce the noise. But then you might as well have used larger pixels to start with.
So I think the right strategy is, figure out how much resolution you need and how large a sensor you can afford, and then make the pixels as large as possible at that resolution. This appears to be what Canon has done.
It would make sense to always process the higher resolution image with denoising and demosaicing algorithms and then save that with a much lower resolution as there is less and less information the lower photon counts you use per pixel.
Otherwise you're just saving noise.
Probably if you did a study from people's online photo albums where they just upload originals without thinking, you'd discover that something like 90% of space is wasted on pure noise. Things like the sky or a smooth constant color area that can not be compressed by the jpg algorithm since it is filled with high frequency noise. (Often of the color variety no less.)
There are limits to sensor efficiency. Photon noise statistics gets worse with smaller absolute aperture lenses (focal length times f-number) and smaller field of view (zoom).
If you're looking for a 20% improvement from 5 MP to 6 MP, you won't find it. It's really more like a 10% improvement. How many MP does a camera have, if it has twice the resolution of a 5 MP camera? It has 20 MP, not 10.
There are also physical limits. The angular resolution depends on the lens aperture size. Smaller cameras have smaller lenses with smaller apertures, which must have lower angular resolution just because of the laws of physics. At the same time, engineering trade-offs mean that smaller lenses usually have larger apertures proportional to their focal length, so this difference is not quite proportional.
But these small cameras do not have unreasonable resolutions. I just looked up the specs for a tiny point-and-shoot, the Nikon Coolpix AW100. From the specs (5-25mm lens, 28-140mm equivalent) we can figure out that the sensor is probably about 6.2mm across (36mm * 5mm / 28mm). The lens is f/3.9, which gives a resolution of about 400 lp/mm in visible light, or about 5000 pixels across 6.2mm. The sensor is 16 MP with about 4600 pixels across it. So the sensor is close to the limits of the optical system.
If you're having trouble noticing improvement above 5 MP, it is probably due to technique rather than equipment.
* You need to use a tripod. (Or a flash. This should be apparent.)
* You must not open the lens fully, this will blur the image from lens imperfections. (Look at the MTF charts of your favorite lens. They're way worse when the lens is fully open.)
* You must not close the lens more than a couple stops, this will blur the image from diffraction of light through the lens aperture. (Rayleigh limit. You can do the math yourself.)
* You need to be very careful with focus. (In my experience, the "hyperfocal distance" is total bunk when you're doing critical work, which makes sense because the hyperfocal formula is based on the assumption that diffraction is irrelevant. When you want maximum sharpness, you'll be shooting near the diffraction limit so it will be relevant, and therefore the hyperfocal formula is wrong. If I remember right, the adjusted formula is based on lens travel distance from farthest to closest subject point, but I've been using the same lens for so long I use intuition.)
>The lens is f/3.9, which gives a resolution of about 400 lp/mm in visible light
Assuming that you've got a perfect lens. As you say yourself "You must not open the lens fully", in which case the f/3.9 is irrelevant.
And even stopped-down, the glass (of which there's several elements in the average compact's zoom) is rarely of the highest quality. All of this makes for an effective resolution some way short of the figures you're quoting.
The quality of the lens can make a huge difference. On my main DSLR, the difference in quality between, say, the 50mm f1.2L and and an identical shot taken with a bog standard 17-85 f/4, even at the same f-stop is vast. If the cheaper lenses on DLSRs aren't able to really give enough quality to fully take advantage of higher MP, I'm not quite sure why you'd think that the ones on a compact would be.
> If the cheaper lenses on DLSRs aren't able to really give enough quality to fully take advantage of higher MP, I'm not quite sure why you'd think that the ones on a compact would be.
You might be surprised at the quality of some of the P&S lenses. There have been excellent lenses on cheap fixed-lens P&S cameras for decades: a modern example is the Sony Cyber-shot DSC-RX100, an older example is the Yahsica Electro series from the 1960s. From resolution tests you can find modern P&S cameras with soft lenses and you can also find ones with real, honest 10+ MP resolutions. You don't need to trust my math, you can go check out test shots online.
> As you say yourself "You must not open the lens fully", in which case the f/3.9 is irrelevant.
I also wasn't really explaining things fully with the maxim. Opening the lens reduces diffraction and increases aberrations. A few good lenses will open all the way to, say, f/2 and still get sharper as you open them up. Some terrible lenses aren't sharp at f/8. However, the trade-off is different at different system sizes. For a P&S, wide open might be the best way to go. For MF and LF, you'll probably want f/11 or f/22 to start with, as ridiculous as that sounds. Of course, this assumes you want maximum resolution.
>You might be surprised at the quality of some of the P&S lenses.
As I have owned several P&S cameras, and done plenty of test shots with them (including in controlled conditions, on tripods in good light etc - despite this not being how the average user is taking the vast majority of their photos), I would not be surprised by the quality of most of them at all. They simply aren't good enough to take a great deal of advantage of much about about 5MP. Yes, you can get ones with pretty good lenses on. But that's far from the norm - and the norm has for years been to fit high megapixel sensors on cameras than can simply not do them justice. Which was the point of my comment.
1. Noise increases as pixel size increases. So you can't just make the pixels bigger.
2. The 5 megapixel camera on an iPhone 4 is pretty amazing. I wouldn't call it a waste of resources.
3. Displacing MF from high-end camera manufacturers requires tiny pixels. You can shoot pictures for e.g. magazines on a $2,000 D600 today, but in 2003 you would have needed at least $4,000 MF camera plus a $10,000+ scanner. Yes, this is just a niche of the photography market, but it's an important, high-profile niche.
In my mind, the quality-conscious consumers were taken by digital at 6 MP, and the studios were taken by digital at 24 MP. There are a few morons around with more money than smarts who will buy crappy cameras because the numbers on the box are higher, but it's clear to me that they're not driving the market right now.
It is simply not true that noise increases as pixel size increases. The exact opposite is widely known to be true. See this page, for example: http://firmitas.org/Sensor.html
I was actually just paraphrasing the linked page from Canon:
> In addition, the sensor's pixels and readout circuitry employ new technologies that reduce noise, which tends to increase as pixel size increases.
The linked page is pretty sketchy, it mentions lenses but not optics. Optical resolution will be limited regardless of whether the lenses are perfectly polished. It's hardly irrelevant too, with modern optical resolution close to sensor resolution. The article makes assertions but no citations or measurements. Worthless junk.
Well they're describing the sensor, not optics, so that part is forgivable.
Also, they're describing a 35mm sensor. This thing will work with existing 35mm film optics, and (apart from physical properties) that's a big part of why the sensor size is interesting.
Would be nice if there were more technical details, but I guess that's press releases for you.
On top of what others have said, let's not forget thermal noise - I can't find it right now but I remember a beercooler-as-SLR-cooler hack someone used to near-freeze his camera, making his long exposure astronomy pictures much less noisy. Also, I've heard that because of this it can be better to shoot a lot of medium-exposure pictures in burst mode and additional-blend them afterwards than having the sensor capture for, say, ten minutes non-stop.
So, since the heatigng up in turn depends largely on current running through the sensor: how does that scale with larger/smaller pixels?
>1. Noise increases as pixel size increases. So you can't just make the pixels bigger.
No, it does not. It also depends on the sensor technology. Newer APS-C 16 and 20 mp cameras have less noise than 3mp APS-C cameras of 10 years ago, even without applying in-camera noise reduction.
No, it's the reverse -- relative to photon collection efficiency, noise decreases as pixel size increases. For a given pixel size, you have various noise sources, and you have the photon collection efficiency of the sensor. As the sensor becomes larger, the photon collection efficiency increases much faster than the noise does.
Quote: "The size of a pixel directly impacts how much noise an image will have in low light, and in some cases even in daylight. The bigger the pixel is, the lower the noise because more photons can reach a bigger pixel sensor."
>Too bad the whole camera business wasted a huge amount of resouces over ten years on a stupid tinier and tinier pixels race.
Not that stupid. While more pixels = more noise holds true for the SAME sensor, it doesn't hold true for later sensors that have better less noise and better gain at the CCD/CMOS level.
Current 16 megapixel cameras have LESS noise and better color rendition than 3mp cameras with the same sensor size 10 years ago.
The example of the burning incense doesn't look it exceeds the threshold around the ember while simultaneously showing the man's face as lit, so I'd assume you are right about the increased dynamic range.
I wouldn't mind a better bit of info about the sensor! For instance I'm guessing its a bayer filtered job, to keep the sensativity (at the expense of sharpness of say Sigma Foveon).
But I can't believe they have just had such an improvement by switching to larger wells and more sensativity, do they have some cunning time domain based filtering too. One advantage digital sensors have is that they can exclude certain bits of noise by merging the image results over time, obviously this comes at the expense of sharpness.
For comparison, the Nikon D4 (a digital SLR) can shoot video at ISO 204,800, and it can resolve details in the dark beyond the ability of the eye, and I suspect this new sensor is more sensitive — or, at least, far more noise free — than that.
EDIT: Judging by some photos I've taken of the night sky, with an f/2.8 lens and with a shutter speed of 1/50s, you need an ISO of around 300,000 to get enough brightness.
If only Stanley Kubrick had this to play with. Instead, he used a f/0.7 lens to shoot some scenes in Barry Lyndon by candlelight, way back in 1975.
Didn't he use the same lenses used by NASA on the cameras for the Apollo missions? I think the larger aperture gave better low-light exposure during the candlelit scene (and may have inadvertantly fuelled conspiracy nuts that think the moon landings were a hoax because they think the sunlight reflected off the regolith was a secondary light source).
Yes, an f/0.7 aperture is a ridiculously wide aperture (thus, it lets in a lot of light[1]), two stops brighter than f/1.4 (a "stop" means a doubling of the amount of light; f/1.2 lenses are the fastest lenses commonly available, and they'll cost around $2k), or four stops faster than f/2.8 (which is what the fastest commonly available zoom lenses are, and they're expensive too). Leica make a very rare f/0.95 lens that would set you back $11k, and it's appropriately called Noctilux.
As for conspiracy theories, what makes it even better is that Stanley Kubrick also made 2001: A Space Odyssey in 1968, which includes some scenes set on the moon, and I've heard of conspiracy theories about how these were used for the "real" moon landing too… (And yes, according to Wikipedia, Kubrick bought three of the ten copies of this lens made, with NASA buying six and Carl Zeiss keeping the other.)
Holy cow! I just visited arethuza's Wikipedia link below...
I read the NASA trivia about the movie on a forum a few years back and it just popped into my memory. I had no idea these were incredibly rare. Yes, the irony him shooting 2001 wasn't lost on me.
Without reflections, space is pretty dark (excluding starlight) so I imagine the new Canon lense will see a lot of space use. Provided it can be made sufficienty RAD-hard.
Distant from stars like our sun, the temperature of space is just 2.73 Kelvins. It's pretty easy to work out the temperature for any given location in space -- just add up the effect of any nearby stars, whose radiation decreases as the square of distance, and add the background temperature of space itself, caused by the Big Bang afterglow.
What this means is that, unless you're close to a star, it is very, very cold. Even at Pluto's relatively nearby location, the average temperature is only 33 Kelvins.
There's also a discontinued Canon f/1.0 that some rental places still have. I shot some stuff with it, and it's pretty amazing to play around with. Even with that though, we had to expose for at least 4 seconds at ISO 1600 to get legit star exposures. I've always fantasized about being able to shoot handheld video of the Milky Way, I'm happy to see that the technology now exists!
The lens in the photo looks like a 50mm f/1.2 (I can't quite make it out in the vid), and I'd be very surprised if they weren't shooting wide open for this demo.
If you are in Los Angeles, you can go to the Stanley Kubrick exhibit at LACMA and see the f/0.7 lens in person, right now. For that matter the entire exhibit is awesome.
Is this a color sensor? The high end astrophotography sensors tend to be monochromatic, with a filter wheel placed in front of the sensor. That way, each well has 3 times as much chance of seeing a photon.
All sensors are monochrome just with red, green, and blue filters over pixels in what's called a Bayer pattern. The only real exception to this rule that I can think of is the Foveon sensor.
of course we would, but photographers wouldn't be too impressed with only capturing 1920(?)x1080 resolution.
This gains it's light gathering abilities by simply having much larger "pixels" on the sensor, which can collect more photons. This is the same principal used on the highest end canon eos body, the 1DX, which has slightly lower resolution than the 5DmkIII in order to gain sensitivity.
Native resolution of this baby looks to be 864 mm^2 / 19 micron^2 ~= 45MPix, in line with the latest Nikon and with rumors of high-density FF Canon. The innovation here is noise reduction and not the sensor itself... obviously Canon will charge an arm an a leg for that video camera - and won't ever make it into a DSLR to not cannibalize their profits.
Why do you say that? Their SLR line already has video cameras in its lineup. It's actually becoming a pretty big thing to use an SLR for a video camera instead of a regular video camera. If that lens was put inside their next EOS 1D or 5D, you'd better believe there'd be a considerable line of people wanting this camera, especially if you could shoot in 1080p or higher, as well as 45MP (or even mid-30's).
Cannibalizing? Not even close. Just the next generation of cameras! They did the same thing between the ... I think it was EOS 50D and EOS 60D
I think it's possible you might be unfamiliar with just how sensitive current sensors are, for photographic use? You can easily capture these kinds of images in very high resolution with a very affordable DSLR body today. The trick is capturing them at only 1080P or 2K, over and over, 24 to 60 times a second, consistently. That's what would be a game changer.
I always find it depressing when I take a picture by moonlight and it ends up looking very much like a picture taken by sunlight, as opposed to looking like the scene looks to our eyes. Always have to keep in mind that a CCD != rods and cones.
Sensitivity is nice but as far as I'm concerned, dynamic range is the main thing that's needed in digital photography. DSLR sensors have gotten much, much better over the last 10 years but there's still plenty of room for improvement.
You're right, about photography. In video, however, you need tons and tons and tons and tons of sensitivity in addition to dynamic range because you can only use fast "shutter" (electronic, not physical, of course) speeds. Needing to create a perfect image 24, 30, or 60 times a second in the dark requires a very different bag of tricks. This is why, for example, having fewer and larger pixels is vastly preferable for motion versus stills.
That's a result of how metering works - it assumes you want a brightly lit scene. You'll either need to use manual exposure or use one of the spot metering modes, where it measures only the brightness of a focal point rather than the overall scene brightness and tries to compensate.
But yes, the metering software in a camera does not know how the scene "should" look like. It can only assume that you want a "properly" exposed photo - such that both under- and overexposure is minimized. If you want something else, you'll have to tell the camera what you want.
Are you asking if it could be used for slow motion capture? Because if you're doing something that could be captured at normal 24/48/60 fps, then I don't see any reason why this camera would care if there was motion or not
I wonder what the anticipated applications are? Surveillance seems like one obvious target (and I guess that's what the moonlight-scene was trying to show), at least if the sensor (and thus the final products) is not ridiculously expensive. Also the fact that it is being demoed at "SECURITY SHOW 2013" points towards surveillance.
As a filmmaker, I'm more impressed by the burning incense. I don't even make live-action video (I do animation), and I STILL desperately want a camera that can do that.
I suspect that the results of showing that video to a room full of filmmakers can be summed up in the phrase "Shut up and take my money."
To the best of my knowledge, there's no reason not to do that - although I don't do stop-motion, I do performance capture.
(The reason I was drooling over the camera wasn't for any particularly practical use - I wouldn't use it day-to-day. I have on occasion done live-action filmmaking, and I know how INCREDIBLY useful uber-low-light performance like that would be. Seeing the potential for mood and Dogme-style shooting makes me want to pick up a Real Camera again ;) )
I bet it can do some insane slowmo frame rates too in normal lighting conditions. For all they put into this sensor, surprised Canon produced such a crapily edited video.
Let's see what codecs and outputs they record before getting too excited. The sensor in the 5d mk III could be so much more if it recorded to something better. Bandwidth is a killer.
Awesome. I've been waiting a long time for camera makers to do something like this after they got tired of increasing megapixels. Really going to open up new worlds of creativity for nighttime and low-light indoor shooting. Even better - make a b/w version with 3x more sensitivity and better sharpness.
So, with all the critical details here about CCD sensors, you expert guys are telling me that I should not rush out and spend $10 K, $20 K, or whatever for the latest Nikon/Canon single lens reflex with a CCD sensor and a bag of lenses and, instead, keep my old Nikon Nikormat with its bag of Nikon lenses?
Let's see, opening the bag for the first time in a while, I have a Nikkor 200 mm f4, a Nikkor 105 mm f2.5, a Nikkor 55 mm f3.5 macro (good for closeups and general usage) with an extension tube for getting even closer, and a Nikkor 50 mm f1.4. Ah, I guess the f1.4 is for taking shots of the girlfriend in her little cocktail dress sitting at the candle lit romantic dinner table waiting for the rest of the evening! Or maybe I should be sitting at the table taking the picture as she comes in with a frilly apron over her cocktail dress and carrying a serving tray with her hand made Sachertorte mit Schlagobers (whipped cream)?
Also in the bag is a Honeywell strobe light: It uses a 900 Volt dry cell battery if can find one and otherwise just uses the A/C adapter. The recharge time on the A/C adapter is fast.
I suspect I will be able to buy film such as Kodak Kodacolor for a long time, but processing is an issue: I've had Sam's Club start with my Kodacolor 35 mm and return to me a CD with JPG images with 1818 x 1228 pixels which seems to be a little below the resolution of the best of the Nikon equipment. At one time Kodak offered Kodak PhotoCD with much higher resolution.
Any suggestions? What about getting good processing and scanning to JPG of the film? Sounds like my old Nikon equipment is still competitive with newer single lens reflex cameras with CCD sensors and roughly 35 mm image width?
E.g., on resolution Dalsa made a CCD image sensor with 48 million pixels, a light detection rectangle 36 x 48
mm, i.e., much like 35 mm film in my Nikon, with each pixel a square 6 microns on a side, capturing about 55,000 electrons per pixel, and, net, 6 K x 8 K pixels. Sounds good until notice that apparently each pixel is for just one color, red, green, or blue, and they use a 2 x 2 array of four such pixels, one red, one blue, and two green, for one 'color' pixel. So, in such 'color' pixels their sensor is 3 K x 4 K and, thus, not necessarily a lot better than 35 mm film?
I had to Google "Bayer filter" -- the Wikipedia page is good. So, yes, without knowing it, I was describing a Bayer filter.
For video, sure, my Nikormat is only a still camera, but my guess is that the Canon CCD sensor will also get used for still photography. Then a question is, would that chip make a much better 35 mm 'class' still camera than my old Nikormat with, say, just Kodacolor film? Apparently on ASA speed values, the Canon sensor would totally blow away any film.
But on resolution, I was guessing that film such as Kodacolor in a good, old still camera such as my Nikormat might still be close to competitive.
But I have fallen so behind in photography that I don't even know what the camera ads mean when they say how many pixels their sensor has. For the Dalsa sensor I described, would the ads say 48 million pixels or just 1/4th of that many, 12 million, after the Bayer filtering?
It's a fantasy image that would be good to photograph, if it happened.
A Sachertorte is a chocolate cake famous from Hotel Sacher in Vienna. It has lots of chocolate and eggs, some apricot jam, and only a little flour. The cake is chocolate, and there is a covering of a soft candy-like chocolate over the whole cake. As is common with coffee and deserts in Vienna, it is commonly eaten with whipped cream.
If you are having a romantic candle lit dinner with a pretty girl wearing a pretty cocktail dress and for desert she goes to the kitchen and comes out wearing a frilly apron over her dress and carrying a tray with that cake and whipped cream, that she made herself just for you, then you have lots of symbols, worth photographing, of a potential of a very happy marriage and home life (although some women would say would be happy mostly only for you) which, of course, is one of the main reasons to take pictures.
> I suspect I will be able to buy film such as Kodak Kodacolor for a long time
Check your suspicions again. Kodak has discontinued, in the past few years:
* Kodachrome
* High color Portra films (the VC versions)
* Ektachrome (pro and consumer)
* Plus-X Pan
I suspect consumer negative film is on the way out. You'll be stuck with Ektar and Portra, or maybe just one of those two choices. All of the above films were discontinued before Kodak filed for Chapter 11 bankruptcy, so perhaps it will only get worse.
THE BIG PROBLEM: You cannot inexpensively scan 35mm at high resolution. Photographers have been saying for years that film is great because you can just hang on to your negatives and wait for the prices on good scanners to come down, but my experience is that the nice scanners ($3,000 Nikon Coolscan 9000) yield an actual resolution much closer to 2000 dpi rather than the advertised 4000 dpi. You could also buy a Flextight or a drum scanner for like $12,000, and you can send good frames out for drum scanning at $40 per frame.
This means that with a good setup you'll be getting roughly 5+ MP pictures out of your 35mm rig.
In most situations, it's not even close to competetive with the quality of a modern 24 MP DSLR with a full frame sensor. It's just a hell of a lot cheaper.
Well, I wrote Kodacolor realizing that Kodak was in Chapter 11 and film is on the way out. But I was hoping that Fuji or maybe something from Europe would be equivalent to Kodacolor, as I recall, called color negative film. But I don't know anything about Fuji film either.
On scanning resolution, thanks for the detail. So, the resolution I am getting now at Sam's Club, 1818 * 1228 = 2,232,504 maybe isn't too awful.
Apparently getting color negative film scanned to something like the old Kodak PhotoCD is still possible at, say,
Their Web site mentions files sizes and file types, but I didn't find much on their resolution in pixels.
So, net, right, it looks like a current, high-end, Nikon or Canon full frame 35 mm digital single lens reflex camera or even a movie camera with the Canon CCD sensor of the OP will be much better than color negative film in my old Nikormat on sensitivity, i.e., ASA 'film speed' and resolution.
So, if I want to dig into photography again, then I should get a sack full of modern equipment and copies of the relevant Adobe software. In the meanwhile, shooting color negative film and getting it scanned at Sam's Club (if they still do that), Wal-Mart, TwoCat, etc. seems like still a relatively good way to take pictures.
This is my today's best Hackernews item, ever. As a Videographer, I've always yearned for better DSLR's that can shoot video under (extreme) low-light conditions. I just hope they launch a consumer/pro-sumer model with this sensor soon! Throw in a low-light lens and you're all set for (ultra) low-light videography!!
97 comments
[ 4.6 ms ] story [ 171 ms ] threadI suppose the actual consumer products will be lighter.
http://www.usa.canon.com/cusa/professional/products/professi...
You can get similarly startling images from a plain point and shoot camera by installing it on a tripod and taking long exposure shots of things. You need 30 to 60 seconds exposure, sometimes available as a "starry sky" scene mode. Architecture is particularly suited as it tends not to move around. :) Even a cheap, small, P&S sensor is much more sensitive to light when you expose for 60 seconds, all while being completely noise-free.
Many sensors are ill-suited to long exposures though. While they can do it, you end up with heat related issues (commonly manifested as stuck pixels and general noise). CCDs also end up saturating and spilling over to adjacent cells, making high dynamic scenes a challenge.
What I'm really wondering about is the dynamic range - wouldn't that also drastically increase?
With the quality of lens on the average compact camera for instance, I struggle to see any real noticable improvement above around 5 MP.
Of course, if you have more resolution, you can reduce the resolution in post-processing, and that will also reduce the noise. But then you might as well have used larger pixels to start with.
So I think the right strategy is, figure out how much resolution you need and how large a sensor you can afford, and then make the pixels as large as possible at that resolution. This appears to be what Canon has done.
Otherwise you're just saving noise.
Probably if you did a study from people's online photo albums where they just upload originals without thinking, you'd discover that something like 90% of space is wasted on pure noise. Things like the sky or a smooth constant color area that can not be compressed by the jpg algorithm since it is filled with high frequency noise. (Often of the color variety no less.)
There are limits to sensor efficiency. Photon noise statistics gets worse with smaller absolute aperture lenses (focal length times f-number) and smaller field of view (zoom).
There are also physical limits. The angular resolution depends on the lens aperture size. Smaller cameras have smaller lenses with smaller apertures, which must have lower angular resolution just because of the laws of physics. At the same time, engineering trade-offs mean that smaller lenses usually have larger apertures proportional to their focal length, so this difference is not quite proportional.
But these small cameras do not have unreasonable resolutions. I just looked up the specs for a tiny point-and-shoot, the Nikon Coolpix AW100. From the specs (5-25mm lens, 28-140mm equivalent) we can figure out that the sensor is probably about 6.2mm across (36mm * 5mm / 28mm). The lens is f/3.9, which gives a resolution of about 400 lp/mm in visible light, or about 5000 pixels across 6.2mm. The sensor is 16 MP with about 4600 pixels across it. So the sensor is close to the limits of the optical system.
If you're having trouble noticing improvement above 5 MP, it is probably due to technique rather than equipment.
* You need to use a tripod. (Or a flash. This should be apparent.)
* You must not open the lens fully, this will blur the image from lens imperfections. (Look at the MTF charts of your favorite lens. They're way worse when the lens is fully open.)
* You must not close the lens more than a couple stops, this will blur the image from diffraction of light through the lens aperture. (Rayleigh limit. You can do the math yourself.)
* You need to be very careful with focus. (In my experience, the "hyperfocal distance" is total bunk when you're doing critical work, which makes sense because the hyperfocal formula is based on the assumption that diffraction is irrelevant. When you want maximum sharpness, you'll be shooting near the diffraction limit so it will be relevant, and therefore the hyperfocal formula is wrong. If I remember right, the adjusted formula is based on lens travel distance from farthest to closest subject point, but I've been using the same lens for so long I use intuition.)
Assuming that you've got a perfect lens. As you say yourself "You must not open the lens fully", in which case the f/3.9 is irrelevant.
And even stopped-down, the glass (of which there's several elements in the average compact's zoom) is rarely of the highest quality. All of this makes for an effective resolution some way short of the figures you're quoting.
The quality of the lens can make a huge difference. On my main DSLR, the difference in quality between, say, the 50mm f1.2L and and an identical shot taken with a bog standard 17-85 f/4, even at the same f-stop is vast. If the cheaper lenses on DLSRs aren't able to really give enough quality to fully take advantage of higher MP, I'm not quite sure why you'd think that the ones on a compact would be.
You might be surprised at the quality of some of the P&S lenses. There have been excellent lenses on cheap fixed-lens P&S cameras for decades: a modern example is the Sony Cyber-shot DSC-RX100, an older example is the Yahsica Electro series from the 1960s. From resolution tests you can find modern P&S cameras with soft lenses and you can also find ones with real, honest 10+ MP resolutions. You don't need to trust my math, you can go check out test shots online.
> As you say yourself "You must not open the lens fully", in which case the f/3.9 is irrelevant.
I also wasn't really explaining things fully with the maxim. Opening the lens reduces diffraction and increases aberrations. A few good lenses will open all the way to, say, f/2 and still get sharper as you open them up. Some terrible lenses aren't sharp at f/8. However, the trade-off is different at different system sizes. For a P&S, wide open might be the best way to go. For MF and LF, you'll probably want f/11 or f/22 to start with, as ridiculous as that sounds. Of course, this assumes you want maximum resolution.
As I have owned several P&S cameras, and done plenty of test shots with them (including in controlled conditions, on tripods in good light etc - despite this not being how the average user is taking the vast majority of their photos), I would not be surprised by the quality of most of them at all. They simply aren't good enough to take a great deal of advantage of much about about 5MP. Yes, you can get ones with pretty good lenses on. But that's far from the norm - and the norm has for years been to fit high megapixel sensors on cameras than can simply not do them justice. Which was the point of my comment.
2. The 5 megapixel camera on an iPhone 4 is pretty amazing. I wouldn't call it a waste of resources.
3. Displacing MF from high-end camera manufacturers requires tiny pixels. You can shoot pictures for e.g. magazines on a $2,000 D600 today, but in 2003 you would have needed at least $4,000 MF camera plus a $10,000+ scanner. Yes, this is just a niche of the photography market, but it's an important, high-profile niche.
In my mind, the quality-conscious consumers were taken by digital at 6 MP, and the studios were taken by digital at 24 MP. There are a few morons around with more money than smarts who will buy crappy cameras because the numbers on the box are higher, but it's clear to me that they're not driving the market right now.
Photon shot noise increases like the square root of the area of the pixel, but the signal will increase like the area.
In that case, the important quantity ( Signal/Noise ) improves like the square root of the area of each pixel.
If readout isn't shot-noise limited, then the payoff for a larger sensor could be as large as the ratio of the areas between old and new.
> In addition, the sensor's pixels and readout circuitry employ new technologies that reduce noise, which tends to increase as pixel size increases.
The linked page is pretty sketchy, it mentions lenses but not optics. Optical resolution will be limited regardless of whether the lenses are perfectly polished. It's hardly irrelevant too, with modern optical resolution close to sensor resolution. The article makes assertions but no citations or measurements. Worthless junk.
Also, they're describing a 35mm sensor. This thing will work with existing 35mm film optics, and (apart from physical properties) that's a big part of why the sensor size is interesting.
Would be nice if there were more technical details, but I guess that's press releases for you.
So, since the heatigng up in turn depends largely on current running through the sensor: how does that scale with larger/smaller pixels?
No, it does not. It also depends on the sensor technology. Newer APS-C 16 and 20 mp cameras have less noise than 3mp APS-C cameras of 10 years ago, even without applying in-camera noise reduction.
No, it's the reverse -- relative to photon collection efficiency, noise decreases as pixel size increases. For a given pixel size, you have various noise sources, and you have the photon collection efficiency of the sensor. As the sensor becomes larger, the photon collection efficiency increases much faster than the noise does.
http://firmitas.org/Sensor.html
Quote: "The size of a pixel directly impacts how much noise an image will have in low light, and in some cases even in daylight. The bigger the pixel is, the lower the noise because more photons can reach a bigger pixel sensor."
Not that stupid. While more pixels = more noise holds true for the SAME sensor, it doesn't hold true for later sensors that have better less noise and better gain at the CCD/CMOS level.
Current 16 megapixel cameras have LESS noise and better color rendition than 3mp cameras with the same sensor size 10 years ago.
But I can't believe they have just had such an improvement by switching to larger wells and more sensativity, do they have some cunning time domain based filtering too. One advantage digital sensors have is that they can exclude certain bits of noise by merging the image results over time, obviously this comes at the expense of sharpness.
Kodak has the patent, so they may have been able to acquire or license it at low cost.
http://web.archive.org/web/20070720002510/http://johncompton...
EDIT: Judging by some photos I've taken of the night sky, with an f/2.8 lens and with a shutter speed of 1/50s, you need an ISO of around 300,000 to get enough brightness.
If only Stanley Kubrick had this to play with. Instead, he used a f/0.7 lens to shoot some scenes in Barry Lyndon by candlelight, way back in 1975.
As for conspiracy theories, what makes it even better is that Stanley Kubrick also made 2001: A Space Odyssey in 1968, which includes some scenes set on the moon, and I've heard of conspiracy theories about how these were used for the "real" moon landing too… (And yes, according to Wikipedia, Kubrick bought three of the ten copies of this lens made, with NASA buying six and Carl Zeiss keeping the other.)
1. http://en.wikipedia.org/wiki/F-number
I read the NASA trivia about the movie on a forum a few years back and it just popped into my memory. I had no idea these were incredibly rare. Yes, the irony him shooting 2001 wasn't lost on me.
Without reflections, space is pretty dark (excluding starlight) so I imagine the new Canon lense will see a lot of space use. Provided it can be made sufficienty RAD-hard.
Lots of infrared too, though.
What this means is that, unless you're close to a star, it is very, very cold. Even at Pluto's relatively nearby location, the average temperature is only 33 Kelvins.
http://en.wikipedia.org/wiki/Cosmic_microwave_background_rad...
They definitely give a cinematic quality to captured video.
http://en.wikipedia.org/wiki/Carl_Zeiss_Planar_50mm_f/0.7
http://www.visual-memory.co.uk/sk/ac/len/page1.htm
and plenty of other fast lenses
http://www.taunusreiter.de/Cameras/Biotar_en.html
and here is a good summary of the developments that went into it (how hard would it be to find this page if not for a link on wiki?!)
http://web.archive.org/web/20090309005033/http://ogiroux.blo...
http://en.wikipedia.org/wiki/Carl_Zeiss_Planar_50mm_f/0.7
This gains it's light gathering abilities by simply having much larger "pixels" on the sensor, which can collect more photons. This is the same principal used on the highest end canon eos body, the 1DX, which has slightly lower resolution than the 5DmkIII in order to gain sensitivity.
> 7.5-times the surface area of the pixels on the CMOS sensor incorporated in Canon's top-of-the-line EOS-1D X and other digital SLR cameras
That puts a 35mm sensor right around HD video resolution, or in the 2-2.1MP range.
>The newly developed CMOS sensor features pixels measuring 19 microns square in size
So there is a possibility that the pixels are merged by the image processor after data is read off the sensor.
Cannibalizing? Not even close. Just the next generation of cameras! They did the same thing between the ... I think it was EOS 50D and EOS 60D
Sensitivity is nice but as far as I'm concerned, dynamic range is the main thing that's needed in digital photography. DSLR sensors have gotten much, much better over the last 10 years but there's still plenty of room for improvement.
But yes, the metering software in a camera does not know how the scene "should" look like. It can only assume that you want a "properly" exposed photo - such that both under- and overexposure is minimized. If you want something else, you'll have to tell the camera what you want.
There's a lot of great information on how to best capture the moon.
I suspect that the results of showing that video to a room full of filmmakers can be summed up in the phrase "Shut up and take my money."
(The reason I was drooling over the camera wasn't for any particularly practical use - I wouldn't use it day-to-day. I have on occasion done live-action filmmaking, and I know how INCREDIBLY useful uber-low-light performance like that would be. Seeing the potential for mood and Dogme-style shooting makes me want to pick up a Real Camera again ;) )
Let's see, opening the bag for the first time in a while, I have a Nikkor 200 mm f4, a Nikkor 105 mm f2.5, a Nikkor 55 mm f3.5 macro (good for closeups and general usage) with an extension tube for getting even closer, and a Nikkor 50 mm f1.4. Ah, I guess the f1.4 is for taking shots of the girlfriend in her little cocktail dress sitting at the candle lit romantic dinner table waiting for the rest of the evening! Or maybe I should be sitting at the table taking the picture as she comes in with a frilly apron over her cocktail dress and carrying a serving tray with her hand made Sachertorte mit Schlagobers (whipped cream)?
Also in the bag is a Honeywell strobe light: It uses a 900 Volt dry cell battery if can find one and otherwise just uses the A/C adapter. The recharge time on the A/C adapter is fast.
I suspect I will be able to buy film such as Kodak Kodacolor for a long time, but processing is an issue: I've had Sam's Club start with my Kodacolor 35 mm and return to me a CD with JPG images with 1818 x 1228 pixels which seems to be a little below the resolution of the best of the Nikon equipment. At one time Kodak offered Kodak PhotoCD with much higher resolution.
Any suggestions? What about getting good processing and scanning to JPG of the film? Sounds like my old Nikon equipment is still competitive with newer single lens reflex cameras with CCD sensors and roughly 35 mm image width?
E.g., on resolution Dalsa made a CCD image sensor with 48 million pixels, a light detection rectangle 36 x 48 mm, i.e., much like 35 mm film in my Nikon, with each pixel a square 6 microns on a side, capturing about 55,000 electrons per pixel, and, net, 6 K x 8 K pixels. Sounds good until notice that apparently each pixel is for just one color, red, green, or blue, and they use a 2 x 2 array of four such pixels, one red, one blue, and two green, for one 'color' pixel. So, in such 'color' pixels their sensor is 3 K x 4 K and, thus, not necessarily a lot better than 35 mm film?
Where am I going wrong?
* practicality of digital vs analog, people will trade quality for convenience
* with Bayer filter you don't lose exactly 1/2 of resolution, the algoritms employed do a little better than that
I had to Google "Bayer filter" -- the Wikipedia page is good. So, yes, without knowing it, I was describing a Bayer filter.
For video, sure, my Nikormat is only a still camera, but my guess is that the Canon CCD sensor will also get used for still photography. Then a question is, would that chip make a much better 35 mm 'class' still camera than my old Nikormat with, say, just Kodacolor film? Apparently on ASA speed values, the Canon sensor would totally blow away any film.
But on resolution, I was guessing that film such as Kodacolor in a good, old still camera such as my Nikormat might still be close to competitive.
But I have fallen so behind in photography that I don't even know what the camera ads mean when they say how many pixels their sensor has. For the Dalsa sensor I described, would the ads say 48 million pixels or just 1/4th of that many, 12 million, after the Bayer filtering?
What am I reading
A Sachertorte is a chocolate cake famous from Hotel Sacher in Vienna. It has lots of chocolate and eggs, some apricot jam, and only a little flour. The cake is chocolate, and there is a covering of a soft candy-like chocolate over the whole cake. As is common with coffee and deserts in Vienna, it is commonly eaten with whipped cream.
If you are having a romantic candle lit dinner with a pretty girl wearing a pretty cocktail dress and for desert she goes to the kitchen and comes out wearing a frilly apron over her dress and carrying a tray with that cake and whipped cream, that she made herself just for you, then you have lots of symbols, worth photographing, of a potential of a very happy marriage and home life (although some women would say would be happy mostly only for you) which, of course, is one of the main reasons to take pictures.
Check your suspicions again. Kodak has discontinued, in the past few years:
* Kodachrome
* High color Portra films (the VC versions)
* Ektachrome (pro and consumer)
* Plus-X Pan
I suspect consumer negative film is on the way out. You'll be stuck with Ektar and Portra, or maybe just one of those two choices. All of the above films were discontinued before Kodak filed for Chapter 11 bankruptcy, so perhaps it will only get worse.
THE BIG PROBLEM: You cannot inexpensively scan 35mm at high resolution. Photographers have been saying for years that film is great because you can just hang on to your negatives and wait for the prices on good scanners to come down, but my experience is that the nice scanners ($3,000 Nikon Coolscan 9000) yield an actual resolution much closer to 2000 dpi rather than the advertised 4000 dpi. You could also buy a Flextight or a drum scanner for like $12,000, and you can send good frames out for drum scanning at $40 per frame.
This means that with a good setup you'll be getting roughly 5+ MP pictures out of your 35mm rig.
In most situations, it's not even close to competetive with the quality of a modern 24 MP DSLR with a full frame sensor. It's just a hell of a lot cheaper.
On scanning resolution, thanks for the detail. So, the resolution I am getting now at Sam's Club, 1818 * 1228 = 2,232,504 maybe isn't too awful.
Apparently getting color negative film scanned to something like the old Kodak PhotoCD is still possible at, say,
http://www.twocatdigital.com/scanning.html
Their Web site mentions files sizes and file types, but I didn't find much on their resolution in pixels.
So, net, right, it looks like a current, high-end, Nikon or Canon full frame 35 mm digital single lens reflex camera or even a movie camera with the Canon CCD sensor of the OP will be much better than color negative film in my old Nikormat on sensitivity, i.e., ASA 'film speed' and resolution.
So, if I want to dig into photography again, then I should get a sack full of modern equipment and copies of the relevant Adobe software. In the meanwhile, shooting color negative film and getting it scanned at Sam's Club (if they still do that), Wal-Mart, TwoCat, etc. seems like still a relatively good way to take pictures.