Slashdot Mirror
Nokia Puts 41MPixel Camera In a (Symbian) Phone
judgecorp writes "We aren't sure what's the strangest thing about Nokia's new offering, the fact that it's got a 41 Megapixel camera or the fact that it runs Symbian. It has a very high resolution sensor and uses oversampling, apparently producing good results in low light. Users can either save a maximum of 38Mpixels, or else zoom and crop for normal resolution images. Observers expected a maximum of one more Symbian phone before Nokia shifts over to Windows Phone. This suggests either a longer life for Symbian — or maybe [that] Symbian was just an easier platform to make a show-stopping device that may turn out to be more of a concept phone."
Unless it has a DSLR-type lens, the limitation is going to be optics, not resolution.
-taktoa
"This suggests either a longer life for Symbian — or maybe Symbian was just an easier platform to make a show-stopping device that may turn out to be more of a concept phone"
Or perhaps the phone has been in development for some time, maybe it takes longer than Marketing announcement cycles to design and deliver new technology.
"The weirdest thing about a mind, is that every answer that you find, is the basis of a brand new cliche" -
Your average phone has a ~4 mm (diameter) lens. This yields an Airy disc of some 1.15 minutes of arc.
Even at a wide field of view (say, 60 degrees), this yields a maximum lateral resolution of some 3200 pixels. Isn't thus any camera with more than ~10 MPixels diffraction limited by the tiny lens, and not sensor limited?
Support a Europe-related section on Slashdot!
I usually don't recommend anything over 10-12MP unless you're going to be blowing up an image to poster-sized. I still use a 6MP camera and it's more than sufficient for daily use. I would much rather have a better sensor since I'm still reducing the image size anyway at 6MP.
I think the big issue is that the camera manufacturers pushed higher MP but never got around to telling Joe Public what exactly MP means to them. Sort of like Intel and AMD pushing faster clock speeds, but when max clock speed reached a plateau in the 3.6-4GHz range they didn't tell consumers a 2GHz quad core with a large cache will likely kill a 3.6GHz single core with a tiny cache so many consumers still go by clock speed alone.
I strongly recommend reading the white paper:
http://europe.nokia.com/PRODUCT_METADATA_0/Products/Phones/8000-series/808/Nokia808PureView_Whitepaper.pdf
I'll look forward to getting 20 of those pictures in an email. Thanks mum, the 10 gig of pictures with nothing but the food you ordered whilst on holiday are great.Oh - I can see some bugs you missed in the salad :) / fires up Photoshop - Mwwhahahhaaa.
The captured image will occupy a small space in the upper left of the picture, the rest will be solid white but when you open the file it will still be 41 million pixels.
Sensor is much larger than a traditional 5MP phone cam sensor:
http://www.dpreview.com/news/2012/02/27/Nokia-808-PureView-with-41MP-sensor
This suggests either a longer life for Symbian - or maybe Symbian was just an easier platform to make a show-stopping device that may turn out to be more of a concept phone.
Or as most of us have figured out, Nokia has been a rudderless company and this is probably the work of the "let's turn Symbian into a smart phone" faction and this is just to recover a little bit of all the money they've wasted, just like the pathetically few N-series phones they released. They probably jumped on the wrong ship when they went all in on Windows Phone, but at least that one is going somewhere. Nokia never managed to agree on one thing and then actually do it well, so Apple and Google ate them for lunch. Epic management fail, if you ask me.
Live today, because you never know what tomorrow brings
I usually don't recommend anything over 10-12MP unless you're going to be blowing up an image to poster-sized. I still use a 6MP camera and it's more than sufficient for daily use. I would much rather have a better sensor since I'm still reducing the image size anyway at 6MP.
Even at 10-12MP you're fine for poster resolutions. This is the resolution current DSLRs operate at (even most full frame ones), simply because it's where you're going to get decent levels of sensitivity in the pixels and not too much noise.
More so, because the lenses, even on DSLRs can't actually resolve that resolution except in absolutely perfect conditions.
Nokia understands it. They have a whitepaper on the technology which explains the use of the chip. Mainly, it is used for digital zooming.
Link to whitepaper: http://europe.nokia.com/PRODUCT_METADATA_0/Products/Phones/8000-series/808/Nokia808PureView_Whitepaper.pdf
Some sample pics, apparently:
http://cdn.conversations.nokia.com.s3.amazonaws.com/wp-content/uploads/2012/02/Archive2.zip
They look OK, and amazing for a phone.
The blog here: http://www.knowyourmobile.com/blog/1263008/nokia_808_pureview_photo_samples_released.html has bot a brief explanation of how the pixels are used and some sample images. Same images as in the zip file in the previous post.
You really didn't even pay attention to the summary, let alone the article did you? The core use here is for super-sampling with dedicated hardware that produces superior 5MP & 8MP images. So... they agree with you! They have created a better sensor. It just so happens that you can also use it in non-super-sampling mode if you really really must.
The core use here is for super-sampling with dedicated hardware that produces superior 5MP & 8MP images.
But you can also "super sample" by making fewer, larger pixels that will collect more light each. Canon stepped back to a 10MP sensor for low-light performance in their G and S9x series cameras (they've since gone back up to 12).
W..w..W - Willy Waterloo washes Warren Wiggins who is washing Waldo Woo.
Keep making fun of Nokia ignorant yankees. One of the main reasons Nokia is non-existent in US is because it tried to stand up to the telcos and protect consumer's rights by not crippling the phones as per the request of your greedy-ass cellular carriers. I guess it won't be making that mistake anymore.
The 808 just goes to show that some companies still employ engineers instead of designers. I mean, Apple has to rip off that patented technology from somewhere. ( http://www.zdnet.com/blog/btl/apple-pays-up-licenses-patents-from-nokia/50558 )
I'm not new here, so I know it's a lot to ask, but in addition to reading the fucking article, I encourage everyone to read the white paper too: http://europe.nokia.com/PRODUCT_METADATA_0/Products/Phones/8000-series/808/Nokia808PureView_Whitepaper.pdf
Also check out the sound quality of the 808 recording (listen with good headphones or speakers to really appreciate the difference) http://www.youtube.com/watch?feature=player_embedded&v=EbLFtF50y9A
The point is not to take 38 megapixel images. I don't know why everyone is focusing on the megapixel, that is not the story here.
The story here is the approach they take, 41 megapixel oversampled images processed algorithmically to produce superior 5 mega pixel images. The story may even be Symbian, definitely not the 41 mega pixel sensor.
Mod up parent and reduce the rest of the comments into nothingness.
Pictures are 5MP standard.
On the short end, combining pixels help to reduce lens abberations due to pixel size.
On the long end, placement of relevant pixels in the centre reduces lens abberations.
And it could not be done in WP7, as the processing power is simply missing there.
Obliged http://xkcd.com/1014/
Actually since this is a near diffraction limited lens working at f/2.4 the spot size is going to be about 0.56um * 2.4 ~ 1.344um on the focal plane. The cycle size is about double, or 2.688um.
Considering it uses a Bayer array, and the pixels are spaced at 1.4um, the green pixels will be spaced at 2um (minimum distance to next green pixel). To properly sample you need at least 2 pixels per cycle (said Mr. Nyquist), but since pixels are not exactly points (they have an area) astronomers working in diffraction limited imaging advise 3x sampling in practice.
What this means is you would need a pixel size of 2.688/3/sqrt(2) ~ 0.63um (or 0.9um if using a Foven-style sensor) to properly sample this lens. 1.4um vastly undersamples the lens, as can be seen near the central area in the available samples: they are razor sharp in the central area, and otherwise are limited by aberrations.
A practical article describing this, with example images, can be seen here:
http://samirkharusi.net/sampling_saturn.html
You cannot proceed from the informal to formal by formal means
Actually, they are. (Nikon thinks so, at any rate, since they just made a DSLR with that kind of resolution.)
What you should ask is "Are these lenses capable of delivering a MTF significantly different from zero at a frequency of 5000 line pairs per picture height?" (In engineering terms, this is 2500 cycles per picture height.) This is an unambiguous criterion for being able to make use of that extra resolution: can the lens deliver detail up to the Nyquist frequency of the sensor?
The answer is pretty unequivocally yes, as can be shown by several tests. The simplest of them is to consider teleconverters. I don't have a 48MP camera handy, but I *do* have a 12MP camera and a 2x teleconverter. Any lens capable of resolving 3000 lp/ph well with a 2x teleconverter will be capable of resolving 6000 lp/ph without it, since an ideal teleconverter just magnifies the image. I have a $125 lens (the Zuiko Digital 35mm f/3.5 macro) which easily satisfies this criterion: it delivers images that contain detail all the way down to the pixel level on a 12MP sensor (i.e. 3000 lp/ph) with a 2x teleconverter. (I use this in the field to get more working distance for bug shots since 35mm is quite short for a macro lens.) Thus, it will deliver useful detail to twice that resolution without the teleconverter, which corresponds to a 48MP sensor at the 4:3 aspect ratio.
This is a $125 lens. (It's also the sharpest I own.) If a cheap lens can do this -- and can do it when focused well into the macro range, which provides special optical challenges -- then there are plenty of more expensive ones that can, too.
Another test is to look at an extremely dense sensor already in existence: the 24MP Sony NEX-7 sensor, which has a crop factor of 1.5x. It's got the same pixel density as a 50+MP fullframe sensor. But people use fullframe lenses on it, such as the Carl Zeiss ones built for the Sony mount, or Leica M lenses, or whatever, and get useful detail out of all that resolution. Those lenses, then, are capable of delivering useful detail to a 50+MP fullframe sensor.
Zoom lenses may have trouble resolving this sort of detail, but even most cheap prime lenses can. (Of course, the very fast ones won't be able to at f/1.4...)
"digital zooming" is just in camera cropping. If your sensor is exceeding the resolving power of your lens your digital zoom is just going to be blurry pixels. You can put a nice big number on the box though.
Silly sensor resolution is silly whether you use it for fake zooming or not.
Honestly, sell your old Canon Rebel and get a Rebel t2i and get a camera that sees in the dark and kicks the crud out of that paleolithic age camera... IF.....
The Rebel t2i is very nice, have one myself. Just don't be cheap on the lens and filters. And if you install "Magic Lantern"(http://magiclantern.wikia.com/wiki/Magic_Lantern_Firmware_Wiki), you really have a nice camera!
And it could not be done in WP7, as the processing power is simply missing there.
Don't they use a dedicated chip to process the images? A driver for it can probably be integrated, given some time; Elop seemed to indicate it may eventually happen. Mind you, the technology has been 5 years in development as it is.
My exception safety is -fno-exceptions.
You can't really compare a cell phone image sensor with a reasonable camera image sensor. There is generally a very significant difference in size between the two. Larger sensor collects more light. More light on the sensor means less sensitivity gain, means less noise in the picture it produces (particularly in low-light). Even if you compare two identically sized CCDs, one with eight times the pixels, you end up with eight pixels with significantly more noise than your single pixel simply because less than 1/8th the light is hitting each of those sensors (hence an even *higher* gain on each of those).
I'm not saying it's going to be a bad cell phone camera, I'm simply saying that increasing the number of pixels over the same surface area of the sensor doesn't in and of itself give you a better image. I think the main benefits they will get with the high resolution CCD here are the ability to do significant digital zooms by cropping and using pixel averaging to keep a reasonable quality for images that aren't zoomed in.
+1 Disagree
You really didn't even pay attention to the summary, let alone the article did you? The core use here is for super-sampling with dedicated hardware that produces superior 5MP & 8MP images. So... they agree with you! They have created a better sensor. It just so happens that you can also use it in non-super-sampling mode if you really really must.
Uh, what?
They're not super sampling, they're down sampling. The physical sensor is always 41 MP. They can only super sample temporally, which is retarded.
They've got a 41 MP sensor. You can run at 41 MP and get the noisiest image in the world. Or you can run at a few MP and get a that 41 MP sensor's image shrunk down, or (even dumber) you can get a crop of that 41 MP sensor (YAY DIGITAL ZOOM!). Super sampling is the process of taking more samples than required for the output. The sensor is always 41 MP and always takes that many samples. Digital zoom throws most of them away. Taking a lower resolution picture just resizes the image. To super sample with a fixed sensor array like in a camera, you'd need to take multiple pictures, either building one larger image and then sizing it down (hope you have a steady hand) or taking multiple pictures over time and laying those on top of each other (hope your subject doesn't move).
The "dedicated hardware" is just some piece of shit DSP that tries to voodoo away the noise. It's gaussian blur + unsharp mask on a chip.