While he's got the ratios right, he's got a tolerance for bad printing that I don't share. VGA for 4 x 6? Maybe if you thought your 110 Instamatic made good enlargements (in those days, 4 x 6 was a premium print). That's newspaper resolution. Double up the dimensions (1.2 MP) and 4 x 6 is just acceptable to my eyes (and it didn't used to be when printer dithering algos weren't what they are today).
Agreed, with a 12 megapixel camera I never felt printing anything over 16 x 20 was acceptable, even with my 22 megapixel 5dmk2 I don't go much larger than 20 x 30.
It all depends on how closely you're viewing them. We have a bunch of photos shot with a 5D (12MP) printed about four feet tall. They look great from a normal viewing distance. Of course, when you put your face six inches away, you can start to see some digital artifacts (stair stepping, etc.) but that's not a problem.
The maximum necessary resolution is related to the intended viewing distance.
"The standard definition of normal visual acuity (20/20 vision) is the ability to resolve a spatial pattern separated by a visual angle of one minute of arc." From that the maximum resolvable dot size can be calculated for a given viewing distance.
20/20 is actually pretty bad eyesight. It's "normal vision", not perfect vision, and when my eyes are "corrected" to 20/20, they're about a quarter-diopter shy of maximal acuity. There is a huge difference to my eyes between pictures printed at a 200-line screen and one printed at a 150-line screen (all other things, like stock and varnish being equal) at a distance of more than one metre (from head-height on a couch to the magazine at the far end of a coffee table). If I were printing from a 12MP original at about four feet tall, I'd certainly run the source through something like Genuine Fractals Print Pro (even a good giclee pattern can't make up for squared-off curves). At four feet in the long dimension (giving a realistic viewing distance of about six feet) I'd notice raggedness at a glance at less than about 80ppi -- but then the dye clouds from a 4x5 negative printed at that size would drive me batty as well.
The one company that seems to have a clue about this is actually Apple. The iPhone 4 has a 5mp camera in it. Some pundits say things like "I wish it had 8 megapixels or more". In doing so, they show their complete ignorance.
Most people couldn't tell the difference between 5 and 8 megapixels. What 8 megapixels buys you is larger files, more noise and slower processing (on a sensor this small). The size of the pixels matter. It's one reason DSLR images tend to be so much better: they have hugely bigger pixels.
I'd love to see more good 5-8mp point and shoot cams rather than these ridiculous 14+ mp noise factories.
I bought the N86 because I realized I never have my camera when I want it, and I always have my cellphone (GPS was nice, but the killer app for me was the 8mp camera).
Canon, Nikon and Panasonic reduced the number of pixels or kept them the same in their high end small sensor compacts. (Canon G11: 10 MP down from 14.7 MP in the predecessor; Canon S90 and S95: 10 MP; Nikon P7000: 10 MP down from 13.5 in the predecessor; Panasonic LX5: the same 10 MP as the 2006 model and a slightly bigger sensor.)
The newest iPhone has a tiny 1/3.2" sensor, those high end compacts usually have 1/1.7" sensors, that’s quite a bit more than double the area. Their pixels are consequently a bit bigger. If Apple made the right decision Canon, Nikon and Panasonic must have, too.
They get it and they know that at least their best informed customers (those who actually buy high end small sensor compacts) will also get it. They are probably only (unnecessarily) afraid of the general public, all those who buy the cheapo point and shoots. If you are willing to pay you get appropriate pixel densities.
I have a Canon S90 and couldn't be happier with it. It had fewer megapixels than several other, cheaper cameras, but use is bearing out that the reviews were correct... they're fewer, but they're better.
You might want to reverse that order:
"If Apple made the right decision Canon, Nikon and Panasonic must have, too."
Canon and Nikon are optics companies. If ~they~ made the right decision, then Apple has weight in their position, too. I don't know enough about Panasonic to state an opinion (I'm neither saying they're bad, nor am I saying their good. I simply know little about them)
The article’s heart is in the right place, but it overstates the case. For example, “Sharpness has very little to do with image quality, and resolution has little to do with sharpness” goes too far if you have a pretty good lens and you’re pretty good at avoiding camera shake.
Pixel count is not nearly as important as marketing makes it out to be, but it is important.
Not only that, but it's a monochrome sensor with a filter wheel. It takes the same shot with multiple filters and the resulting images are merged back on earth.
My friend worked on that project.
[Edit]
It has 12μm x 12μm pixels whereas the 5D has 8μm and the 5DmkII has 6μm pixels. However each pixel on the Canons contain three (RGB) photosites to the rover's one.
Ok, here we're getting into confusion about terminology here.
To clarify:
In the Bayer filtered Canon CCD each individual photosensor is covered in a colored filter [1] in a pattern such as:
B G
G R
So when capturing an image 1/4 of the sensors are capturing blue light, 1/4 capturing red, and 2/4 capturing green.
In contrast the CCD in the rover's cameras do not have filters in front of individual photosensors. Instead, there is a set of filters on a disc which are mechanically positioned in front of the sensor. Using this method three images are taken. On each with a red, green, and blue filter covering 4/4 of those same sensor elements giving you 2x-4x the number of elements capturing each color and thus much higher image quality at a given resolution.
B B
B B
then
G G
G G
then
R R
R R
Similar multi-shot technology was used in the early days of terrestrial digital photography to provide better color, exposure latitude and resolution. [2] for a bit of discussion on multishot vs. single shot medium format digital camera backs.
The part that threw me off was where you said that the Canon sensor contains 3 photosites per pixel -- which based on this pots leads me to believe that you meant that it uses 3 photo sites to record each pixel, which is still not entirely accurate; the Rover's capture every color at every photosite, so they use all of their sensor's resolution. Canon (and everyone else but Foveon, for that matter) making a single-shot sensor can't do that, they instead sacrifice the real resolution of the sensor due to the Bayer filter, and use interpolation to make up the difference.
Not to mention that most of those tiny cameras with huge pixel counts don't have optics good enough to resolve those pixels, so they really are pointless.
I used to work for... A press agency you've all heard of. We had lots of big prints up about the place. Now, some of these were shot on films like Fuji NPZ (ISO 800 colour negative) or first-generation photojournalist DSLRs (e.g. Nikon D1H, 2.1MP). At 60x40", they looked great. Nikon knew this; their second generation was "only" 4MP (D2H).
What matters in resolution is lp/mm; that is how many distinct pairs of black and white lines can fit within 1mm. See here: http://www.bobatkins.com/photography/digital/size_matters.ht... The long and the short of it is, unless you are shooting on a tripod with a cable release and a prime lens, there's really not much point worrying about megapixels. I once had a huge row with people on a mailing list about this; try as I might I couldn't make them understand the simple idea that it doesn't matter how many pixels you have if they can't all be assigned distinct values, there's no additional resolution there. They were the kind of people that bought the most expensive DSLR and stuck the cheapest lens on it. Then again if you'd spent 5 grand on a camera and someone did the maths for you you might be upset too... ;-)
lp/mm is just a unit of measure for resolution and is directly constrained by the number of pixels. You can't resolve 10 lines with less than 20 pixels and in practice you need more.
The article you quoted compares the same resolution in different sized sensors to show that smaller sensors need to resolve more lp/mm in sensor area to achieve the same file in the end. It shows that for the same MP count a larger sensor is better. What we are discussing is the opposite, for the same sensor size MP count is better. And as far as resolution is concerned that tends to be the case, even if the lens is the biggest bottleneck.
In DSLRs we're getting to a point where resolution is more constrained by camera movement and lens quality than number of pixels but at the same time both those bottlenecks are being addressed with modern lens design and anti-shake techniques. We've come a long way in actual practical resolution from the 2MP/4MP cameras you mentioned.
That's very wrong. The number of photons is limited, the effective light-sensitive area drops with more MP, and with small sensors with more MP you then get more noise than a signal. It's hidden by post-processing but that also drops the real information. So less MP is better in a lot of common circumstances.
>And as far as resolution is concerned that tends to be the case
I was only talking about resolution. Lower MP gives you less noise but it won't give you more resolution. In the extreme case where you push the ISO maybe the extra noise can overcome the extra resolution. To make it simple lets say that at base ISO on a sunny day the higher MP will out-resolve the lower MP for a given sensor generation.
My point isn't that low MP isn't good for many applications it's that saying that higher MP doesn't lead to higher resolution is just wrong. If you want resolution you need pixels to produce it and current cameras are mostly not wasting the resolution away.
The excessive MP problem only really exists in consumer compacts. High-end compacts and DSLRs haven't been stretching the limits too much. The highest MP DSLRs right now have 24MP on a full-frame sensor, which is about the same density as 10MP on an APS sensor. Some of the APS ones have pushed that on to ~16MP with accompanying improvements in the individual pixels. Maybe you're thinking of camera phones with 8MP, those may very well be in the extreme case you mention where noise overcomes the extra resolution.
If you exclude professionals, nobody buys "full-frame" sensors. Excluding professionals, people don't even know what that means. Nobody brags about sensor size, everybody brags he has "more MP" than before.
Also almost nobody wants to make photos only in the direct sunlight. Normal conditions are conditions with less than optimal light.
So excluding "full-frame," more MP is actually worse, and people buy simply by believing "bigger number better."
Edit: Let me stress once again: I'm not discussing a scenario of a "professional with a slave to carry all his equipment and another guy to set and hold the lights" but a "normal person" carrying not-too-expensive camera or even his phone and expecting "more MP better pictures."
That's one of the most useful things about dpreview.com, they quote pixels per square cm as well. 2 pixels, one black, one white have more resolution than 10 pixels, all grey.
I was only mentioning full-frame as those are the highest MP DSLRs available. I then mentioned APS cameras, which a lot of people do buy, especially the entry level stuff. In that segment and even in less than perfect light the higher MP cameras have been getting better, not worse.
There's an argument to be made that if the manufacturers had stopped at say 12MP (to pick a number) and put all their effort into noise we would have better cameras now for the kind of use people actually put them through. In fact Nikon did just that at the high-end with the D700 and D3. There isn't however much of an argument to be made that the 16MP APS cameras of today are worse than the 12MP cameras of a couple years back. The latest models seem to have both more pixels and lower noise.
Maybe you'd like a lower MP camera with even lower noise and maybe you can't buy it because MP sells cameras and thus shapes development. That could be a problem, but saying "excluding full-frame, more MP is actually worse" isn't really true. My impression is the complaints about MP are mostly right in compacts and everywhere else tend to be geek snobbery.
> Lower MP gives you less noise but it won't give you more resolution.
This is not a generally true statement (that is, it is sometimes true, but it depends). Sometimes the higher percentage of overhead on the chip required by smaller pixels reduces the overall amount of "actual" information collected. That is, more extra noise is being added than extra signal.
For example, my D300s (12M) at 4288x2848 pixels can be printed at 300dpi (professional photo prints) at a maximum size of 14.3x9.5(inch) = 36.32x24.13(cm) = a little bit smaller than an A3.
The same picture can be printed @150 dpi on an A1 (84.10 cm x 59.40 cm) leaving a little bit of border.
What annoys me a lot is increase in megapixels in camera phones and stagnation in tesktop\notebook screen resolution. I don't need multimegapixel camera in phone. To share photo with friends on facebook or twitter 0.3MP is more then enough but processing of multimegapixel image is not easy task for small ARM processor, so you get very slow response at all phone photo capturing and processing applications and getting annoying delays.
According to desktop screen resolution you are lucky if you can get screen with more then 2MP resolution. And this situation is same for years. It even getting worse after switching to wide screen because now you are getting less pixels on same diagonal and resolution.
You cannot directly compare digital camera megapixels with display megapixels. The former uses Bayer-interpolation (except very few cameras with Foveon sensor) while the latter has true RGB pixels. In the other words, the digital camera's pixel count of two million is shared between R, G, G, B pixels (500k of each – yes, most sensors have double green pixels). A Full HD monitor has 3 * 1920 * 1080 (sub)pixels.
Because of Bayer interpolation, the true resolution of 3MP sensor roughly equals 2MP screen resolution.
(Yes; I fully agree with your point that a phone doesn't need megazillion pixels. Sensor size is more important factor when speaking of image quality. [Not to mention the optics.])
I'm not comparing them directly I'm comparing increase in resolution for camera phones and monitors. Resolution for mainstream camera phone changed from 0.3MP to 5 MP (same Bayer-interpolated megapixels, technology is the same) but I still need to use same 2MP (true megapixels) monitor for years.
34 comments
[ 3.3 ms ] story [ 88.5 ms ] threadThe maximum necessary resolution is related to the intended viewing distance.
"The standard definition of normal visual acuity (20/20 vision) is the ability to resolve a spatial pattern separated by a visual angle of one minute of arc." From that the maximum resolvable dot size can be calculated for a given viewing distance.
http://www.ndt-ed.org/EducationResources/CommunityCollege/Pe...
http://www.clarkvision.com/imagedetail/eye-resolution.html
(Sidenote: billboards usually printed at around 20-30ppi)
http://www.nytimes.com/2007/02/08/technology/08pogue.html?_r...
The one company that seems to have a clue about this is actually Apple. The iPhone 4 has a 5mp camera in it. Some pundits say things like "I wish it had 8 megapixels or more". In doing so, they show their complete ignorance.
Most people couldn't tell the difference between 5 and 8 megapixels. What 8 megapixels buys you is larger files, more noise and slower processing (on a sensor this small). The size of the pixels matter. It's one reason DSLR images tend to be so much better: they have hugely bigger pixels.
I'd love to see more good 5-8mp point and shoot cams rather than these ridiculous 14+ mp noise factories.
Apple still do deserve props for going for a better camera + lens, rather than the samsung route of more rez.
http://europe.nokia.com/find-products/devices/nokia-n86-8mp
I bought the N86 because I realized I never have my camera when I want it, and I always have my cellphone (GPS was nice, but the killer app for me was the 8mp camera).
The newest iPhone has a tiny 1/3.2" sensor, those high end compacts usually have 1/1.7" sensors, that’s quite a bit more than double the area. Their pixels are consequently a bit bigger. If Apple made the right decision Canon, Nikon and Panasonic must have, too.
They get it and they know that at least their best informed customers (those who actually buy high end small sensor compacts) will also get it. They are probably only (unnecessarily) afraid of the general public, all those who buy the cheapo point and shoots. If you are willing to pay you get appropriate pixel densities.
Canon and Nikon are optics companies. If ~they~ made the right decision, then Apple has weight in their position, too. I don't know enough about Panasonic to state an opinion (I'm neither saying they're bad, nor am I saying their good. I simply know little about them)
Pixel count is not nearly as important as marketing makes it out to be, but it is important.
My friend worked on that project.
[Edit]
It has 12μm x 12μm pixels whereas the 5D has 8μm and the 5DmkII has 6μm pixels. However each pixel on the Canons contain three (RGB) photosites to the rover's one.
http://www.mwoa.org/Ch31.pdf
To clarify:
In the Bayer filtered Canon CCD each individual photosensor is covered in a colored filter [1] in a pattern such as:
B G G R
So when capturing an image 1/4 of the sensors are capturing blue light, 1/4 capturing red, and 2/4 capturing green.
In contrast the CCD in the rover's cameras do not have filters in front of individual photosensors. Instead, there is a set of filters on a disc which are mechanically positioned in front of the sensor. Using this method three images are taken. On each with a red, green, and blue filter covering 4/4 of those same sensor elements giving you 2x-4x the number of elements capturing each color and thus much higher image quality at a given resolution.
B B B B then G G G G then R R R R
Similar multi-shot technology was used in the early days of terrestrial digital photography to provide better color, exposure latitude and resolution. [2] for a bit of discussion on multishot vs. single shot medium format digital camera backs.
[1] http://en.wikipedia.org/wiki/Bayer_filter
[2] http://www.luminous-landscape.com/reviews/cameras/h3d50ii.sh...
What matters in resolution is lp/mm; that is how many distinct pairs of black and white lines can fit within 1mm. See here: http://www.bobatkins.com/photography/digital/size_matters.ht... The long and the short of it is, unless you are shooting on a tripod with a cable release and a prime lens, there's really not much point worrying about megapixels. I once had a huge row with people on a mailing list about this; try as I might I couldn't make them understand the simple idea that it doesn't matter how many pixels you have if they can't all be assigned distinct values, there's no additional resolution there. They were the kind of people that bought the most expensive DSLR and stuck the cheapest lens on it. Then again if you'd spent 5 grand on a camera and someone did the maths for you you might be upset too... ;-)
Also note that cinema is only 2000 pixels across.
The article you quoted compares the same resolution in different sized sensors to show that smaller sensors need to resolve more lp/mm in sensor area to achieve the same file in the end. It shows that for the same MP count a larger sensor is better. What we are discussing is the opposite, for the same sensor size MP count is better. And as far as resolution is concerned that tends to be the case, even if the lens is the biggest bottleneck.
In DSLRs we're getting to a point where resolution is more constrained by camera movement and lens quality than number of pixels but at the same time both those bottlenecks are being addressed with modern lens design and anti-shake techniques. We've come a long way in actual practical resolution from the 2MP/4MP cameras you mentioned.
That's very wrong. The number of photons is limited, the effective light-sensitive area drops with more MP, and with small sensors with more MP you then get more noise than a signal. It's hidden by post-processing but that also drops the real information. So less MP is better in a lot of common circumstances.
>And as far as resolution is concerned that tends to be the case
I was only talking about resolution. Lower MP gives you less noise but it won't give you more resolution. In the extreme case where you push the ISO maybe the extra noise can overcome the extra resolution. To make it simple lets say that at base ISO on a sunny day the higher MP will out-resolve the lower MP for a given sensor generation.
My point isn't that low MP isn't good for many applications it's that saying that higher MP doesn't lead to higher resolution is just wrong. If you want resolution you need pixels to produce it and current cameras are mostly not wasting the resolution away.
The excessive MP problem only really exists in consumer compacts. High-end compacts and DSLRs haven't been stretching the limits too much. The highest MP DSLRs right now have 24MP on a full-frame sensor, which is about the same density as 10MP on an APS sensor. Some of the APS ones have pushed that on to ~16MP with accompanying improvements in the individual pixels. Maybe you're thinking of camera phones with 8MP, those may very well be in the extreme case you mention where noise overcomes the extra resolution.
Also almost nobody wants to make photos only in the direct sunlight. Normal conditions are conditions with less than optimal light.
So excluding "full-frame," more MP is actually worse, and people buy simply by believing "bigger number better."
Edit: Let me stress once again: I'm not discussing a scenario of a "professional with a slave to carry all his equipment and another guy to set and hold the lights" but a "normal person" carrying not-too-expensive camera or even his phone and expecting "more MP better pictures."
There's an argument to be made that if the manufacturers had stopped at say 12MP (to pick a number) and put all their effort into noise we would have better cameras now for the kind of use people actually put them through. In fact Nikon did just that at the high-end with the D700 and D3. There isn't however much of an argument to be made that the 16MP APS cameras of today are worse than the 12MP cameras of a couple years back. The latest models seem to have both more pixels and lower noise.
Maybe you'd like a lower MP camera with even lower noise and maybe you can't buy it because MP sells cameras and thus shapes development. That could be a problem, but saying "excluding full-frame, more MP is actually worse" isn't really true. My impression is the complaints about MP are mostly right in compacts and everywhere else tend to be geek snobbery.
This is not a generally true statement (that is, it is sometimes true, but it depends). Sometimes the higher percentage of overhead on the chip required by smaller pixels reduces the overall amount of "actual" information collected. That is, more extra noise is being added than extra signal.
PS: I somtimes notice individual pixels on a still shot in 2k theaters, but there is no way I can do that in the middle of an action shot.
For example, my D300s (12M) at 4288x2848 pixels can be printed at 300dpi (professional photo prints) at a maximum size of 14.3x9.5(inch) = 36.32x24.13(cm) = a little bit smaller than an A3.
The same picture can be printed @150 dpi on an A1 (84.10 cm x 59.40 cm) leaving a little bit of border.
According to desktop screen resolution you are lucky if you can get screen with more then 2MP resolution. And this situation is same for years. It even getting worse after switching to wide screen because now you are getting less pixels on same diagonal and resolution.
Because of Bayer interpolation, the true resolution of 3MP sensor roughly equals 2MP screen resolution.
(Yes; I fully agree with your point that a phone doesn't need megazillion pixels. Sensor size is more important factor when speaking of image quality. [Not to mention the optics.])