Discussion
Just wanted to get peoples thoughts on ISO's....
This is taken at ISO50 (Stop crying that D70 owner at the back!)The compression does it no favours, but I was rather pleased with the smooth gradation.
And this is a 100% crop, no sign of pixilation.
Seems to be a good argument for the "Less is more" side.
Martin.
This is taken at ISO50 (Stop crying that D70 owner at the back!)The compression does it no favours, but I was rather pleased with the smooth gradation.
And this is a 100% crop, no sign of pixilation.
Seems to be a good argument for the "Less is more" side.
Martin.
Hate to tell you Martin, but the "ISO" setting has nothing to do with pixelation. The pixels are the same, all that changes is that the dynamic range of the sensor is compressed to offer viewable pictures with less light.
The downside of that compression is that noise on the electronics gets amplified, and hence you get "noise" in the colours of a picture.
This is different to the grain on a film where the physical size of the molecules that reacted to light got bigger as the ISO got higher, and hence the 'resolution' did actually drop.
The two look similar, but are very different.
For digital "ISO", think of listening to a very quiet voice on an old record player, compared to thrash metal. In order to hear the quiet sound you would have to turn up the volume (narrow the input dynamic range for the same output), and although you could make the quiet voice as "loud" as the thrash metal, there would be a lot of "noise" in the background. If you made the thrash metal the same level as the quiet voice, you wouldn't hear any noise because you had turned the volume down so low that the noise wouldn't be noticable.
ISO50 therefore has less "noise" than ISO800 because the input dynamic range of ISO50 (i.e. the range of light that gives rise to the 0-255 levels of RGB on the output) is much wider than the input range of ISO800.
On the 1d II there is a tradeoff in how it get's to ISO50, and it is likely (although I haven't seen any tests) that the actual dynamic range will be 'truncated'. This doesn't matter as the Canon stuff I've seen says ISO50 is "a tool to grant studio photographers increased control of F-stop settings in flat studio lighting setups" where dynamic range is less of an issue.
If you want the maths - see here http://clarkvision.com/imagedetail/digital.signal.to
Hope that helps!
J
The downside of that compression is that noise on the electronics gets amplified, and hence you get "noise" in the colours of a picture.
This is different to the grain on a film where the physical size of the molecules that reacted to light got bigger as the ISO got higher, and hence the 'resolution' did actually drop.
The two look similar, but are very different.
For digital "ISO", think of listening to a very quiet voice on an old record player, compared to thrash metal. In order to hear the quiet sound you would have to turn up the volume (narrow the input dynamic range for the same output), and although you could make the quiet voice as "loud" as the thrash metal, there would be a lot of "noise" in the background. If you made the thrash metal the same level as the quiet voice, you wouldn't hear any noise because you had turned the volume down so low that the noise wouldn't be noticable.
ISO50 therefore has less "noise" than ISO800 because the input dynamic range of ISO50 (i.e. the range of light that gives rise to the 0-255 levels of RGB on the output) is much wider than the input range of ISO800.
On the 1d II there is a tradeoff in how it get's to ISO50, and it is likely (although I haven't seen any tests) that the actual dynamic range will be 'truncated'. This doesn't matter as the Canon stuff I've seen says ISO50 is "a tool to grant studio photographers increased control of F-stop settings in flat studio lighting setups" where dynamic range is less of an issue.
If you want the maths - see here http://clarkvision.com/imagedetail/digital.signal.to
Hope that helps!
J
simpo two said:
'Suddenly, Martin's favourite selling point evaporated...'
Not evaporated, John, just got kicked into second place. After reading what Stuart (Srider) said about it I had a little read, and came to the conclusion that I should keep my mouth shut untill I know what I'm talking about. Still, a little bit of D70 baiting never killed anyone
Martin.
PS - 1D now set to ISO100, I take your D70 is too?
V6GTO said:
PS - 1D now set to ISO100, I take your D70 is too?
Martin, if I had *ever* had the slightest desire for ISO50, you would have a point. As it is you are simply wearing out your keyboard; you may as well be boasting that it's blue for all it matters to me. Am I interested? No I'm not. Would I use it if I had it? No I wouldn't.
>> Edited by simpo two on Tuesday 22 November 15:26
simpo two said:
PS - 1D now set to ISO100, I take your D70 is too?
Martin, if I had *ever* had the slightest desire for ISO50, you would have a point. As it is you are simply wearing out your keyboard; you may as well be boasting that it's blue for all it matters to me. Am I interested? No I'm not. Would I use it if I had it? No I wouldn't.
>> Edited by simpo two on Tuesday 22 November 15:26
John...see all those smiling/laughing faces on my post? They mean that it was posted lightheartedly, goodhumouredly, tounge-in-cheek and never a thought of boasting. As I said, more of a bit of good natured ribbing, something you have not been averse to yourself in the past.
Martin.
A serious question regarding ISOs - I've frequently in the past shot Kodachrome 25 because of its great colours and image quality (despite the need for extra light) however ignoring the differences between Canons and Nikons, why is it that there doesn't seem to be the need to ISOs on digital cameras of lower than 100/200?
p.s I haven't read the link to the signal/noise article
p.s I haven't read the link to the signal/noise article
V6GTO said:
'Suddenly, Martin's favourite selling point evaporated...'
Not evaporated, John, just got kicked into second place. After reading what Stuart (Srider) said about it I had a little read, and came to the conclusion that I should keep my mouth shut untill I know what I'm talking about. Still, a little bit of D70 baiting never killed anyone
Martin.
PS - 1D now set to ISO100, I take your D70 is too?
Okay, after shooting Martin down in flames, it's time to stick up for him (a bit).
As I mentioned briefly in the other thread, I do use ISO 50 even though it does compromise dynamic range slightly. It's very useful for me when shooting motorsport from behind fences during daylight. In this situation, I need to shoot with the lens close to wide-open, but want a slow shutter speed to blur the wheels on the cars to give a sense of motion. Without going to a very low ISO, the only solution is to add ND filters. Unfortunately, while this does give a slower shutter speed it also hampers the autofocus performance of the camera, since there's actually less light getting to the autofocus sensor. This is exacerbated by the fact that the grey haze that you get from the fence already compromises autofocus.
I'll concede that this is a tenuous point to a lot of people, but it happens to me quite frequently. I also think it's relevant since this is a motoring site.
Now, with that sorted do you Nikon boys want to talk about usable pictures at ISO 3200? No, thought not
joust said:
Hate to tell you Martin, but the "ISO" setting has nothing to do with pixelation. The pixels are the same, all that changes is that the dynamic range of the sensor is compressed to offer viewable pictures with less light.
The downside of that compression is that noise on the electronics gets amplified, and hence you get "noise" in the colours of a picture.
This is different to the grain on a film where the physical size of the molecules that reacted to light got bigger as the ISO got higher, and hence the 'resolution' did actually drop.
The two look similar, but are very different.
For digital "ISO", think of listening to a very quiet voice on an old record player, compared to thrash metal. In order to hear the quiet sound you would have to turn up the volume (narrow the input dynamic range for the same output), and although you could make the quiet voice as "loud" as the thrash metal, there would be a lot of "noise" in the background. If you made the thrash metal the same level as the quiet voice, you wouldn't hear any noise because you had turned the volume down so low that the noise wouldn't be noticable.
ISO50 therefore has less "noise" than ISO800 because the input dynamic range of ISO50 (i.e. the range of light that gives rise to the 0-255 levels of RGB on the output) is much wider than the input range of ISO800.
On the 1d II there is a tradeoff in how it get's to ISO50, and it is likely (although I haven't seen any tests) that the actual dynamic range will be 'truncated'. This doesn't matter as the Canon stuff I've seen says ISO50 is "a tool to grant studio photographers increased control of F-stop settings in flat studio lighting setups" where dynamic range is less of an issue.
If you want the maths - see here http://clarkvision.com/imagedetail/digital.signal.to
Hope that helps!
J
Most of what you say about noise is true, except that you imply that it doesn't reduce the sensor's ability to resolve detail in an image, which isn't true. While the actual number of pixels that are output from the sensor is the same, the quality of the signal goes down as the ISO increases, so effectively you lose resolution and detail.
Noise reduction adds to this effect by blurring the detail to reduce the noise. Take two pictures of the same scene at ISO 100 and ISO 1600 and it's very obvious.
srider said:I'd conceed that the end apparent resolution is lower when you take the effect of everything you say, due to the circle of confusion being larger, but again as you say the underlying sensor resolution is identical.
Most of what you say about noise is true, except that you imply that it doesn't reduce the sensor's ability to resolve detail in an image, which isn't true. While the actual number of pixels that are output from the sensor is the same, the quality of the signal goes down as the ISO increases, so effectively you lose resolution and detail.
Noise reduction adds to this effect by blurring the detail to reduce the noise. Take two pictures of the same scene at ISO 100 and ISO 1600 and it's very obvious.
As Martin's post was about "artifacts", which are the opposite of smoothing to reduce noise, I was merely point out he was talking tosh
You are, indeed, correct, as I am
J
simpo two said:Putting my CMOS VLSI design hat on (I used to design CMOS chips for a living many years ago) that isn't actually correct.
There has to be an optimum nmber of photons to hit each pixel for 100% best results, no? So I'm guessing there's an optimum ISO for each pixel size.
Sticking with Canon as they use CMOS chips the sensitivity of a pixel is determined by many things
1) Its physcial size
2) The inter-pixel spacing (which is dictated by the noise cancelling circuitry next to each "sensor"
3) The substrate (affects charge leakage)
4) The exact makeup of the "Metal Oxide" (the MO of CMOS), and the various metals used.
To compared one CMOS chip with another is not very valid unless you know that stuff. However, you may be suprised that the actual "senor" part of one pixel can be <10% of the actual "pixel" size - the rest of the space is taken up by the supporting infrastructure.
Finally, there are two types (well, three actually) of sensors you can build. You can build logarithmic, linear, or "linlog" sensors where the "output" number either responds linearly, logarthmically, or a combination of the two.
Without knowing all those factors it's hard to say what the sensitivity of the final image array will be....
So there
J
simpo two said:No - as unfortunatly the "higher end" gets compressed and so you loose highlights. It's the reason most developments and patents recently have been trying to combine the two - so you get the benifit of log at the bottom and linear at the top.
I'm guessing 'logarithmic' would give more dynamic range?
simpo two said:They do that for things like X-Ray detectors, but it seriously increases the cost, and I mean by a massive factor.
Plus, why not put the supporting gubbins under the pixel insteda of next to it, so the pixel can be bigger?
A APS-C CCD is a massive chip by fab standards - a P4 is 1/4 of the physical size. Unfortunatly, a wafer will always have imperfections across it, which means that as the size of the chip increases the chance of having an imperection on the chip increases with a square law. This means that the wasteage goes up massivly for large chips.
If you build multi-layer chips then it increases the number of lithography stages, and again there are chances of that going wrong. The bigger the chip, the more you waste.
(imagine one imperfection top tight of a square. For a chip the size of the square you waste 1 whole chip. If you half the size of the chip, you can then get 4 chips in the same area, and hence you waste 1 chip, but get 3 good ones, if you 1/4 the size you get 15 good ones, 1 wasted one etc.)
It's all down to a simple cube law of wastage on the fabrication - as you increase the size/complexity you waste more chips (and the bigger the chip the longer it takes to find out it's a "duff"
and hence the price goes up massivly. I'd imagine the main cost behind the full frame sensor cameras is actually the fab costs - when I moved from 100,000 transistor designs to 400,000 10 years ago, the cost of fabbing the test chips went up by a factor of 200!
J
simpo two said:Fortunatly I worked out you can never make serious money designing chips unless you are lucky, very very lucky. Hence I moved into more traditional lines of making money!
When you make your ultra DR, ultra low-noise chip, remember me
J
joust said:
It's all down to a simple cube law of wastage on the fabrication - as you increase the size/complexity you waste more chips (and the bigger the chip the longer it takes to find out it's a "duff"
Good point. How about stitching 4 smaller ones together?
Gassing Station | Photography & Video | Top of Page | What's New | My Stuff