Optimum ISO (Canon, Fuji, Nikon, Sony)

[EinsteinsGhost]

I recently picked up my back up interchangeable lens camera, a used Sony NEX-3. This camera is rated for ISO range from 200 thru 12800 (and auto, which uses these and interpolated ISOs as necessary). This was also true of its higher trim (NEX-5). With launch of NEX-C3 and NEX-5N, Sony added ISO 100 and ISO 25600 at the extremes (compared to NEX-3 and 5). The top model (NEX-7), however goes from 100 thru 16000 (which is an interesting number in itself).

Besides Sony NEX (3 and 5), I have come across specs for many cameras that have “standard” ISO range from 200 thru 6400 and “expanded” ISO range from 100 thru 12800, or so. Something about that, begs a question around the generally accepted premise that selecting the lowest ISO gets the best result. I never thought that ISO at lower side would be offered under “expanded range”.

One of my conclusions was that, lowest ISO may not be the best bet for best quality (which may also depend on what "quality" entails). So what might be? Should we avoid selecting lowest ISO setting as one normally does with aperture settings on a lens (wide open results are rarely, if at all, sharper than if the lens is stopped down by a stop or two or three)?

This led me to a web search and discovery that some have claimed (and demonstrated) that selecting ISO in multiples of 160 can yield better results than the nearest “rounded down” ISO. I wonder if this applies universally, and if there is a reason to this? It could also mean that in a Canon, ISO160 (if selectable) would produce better results than ISO100. Could that be the idea behind expanded range also including lower ISO rating as opposed to it being standard?

Based on my personal experience with compact (P&S) cameras with tiny sensors, however, is that lesser is greater. But this doesn't seem to apply for cameras with larger sensors.

Any input on the subject?

[Lamplight]

In my experience, lower numbers always seem to produce better results, whether it was with my Digital Rebel, 20D, 7D, or Fuji X10. At least to my eye, anyway.

[Floyd_Davidson]

For the most part your observations and question are pretty astute!

Originally Posted by EinsteinsGhost

I recently picked up my back up interchangeable lens camera, a used Sony NEX-3. This camera is rated for ISO range from 200 thru 12800 (and auto, which uses these and interpolated ISOs as necessary). This was also true of its higher trim (NEX-5). With launch of NEX-C3 and NEX-5N, Sony added ISO 100 and ISO 25600 at the extremes (compared to NEX-3 and 5). The top model (NEX-7), however goes from 100 thru 16000 (which is an interesting number in itself).

Besides Sony NEX (3 and 5), I have come across specs for many cameras that have “standard” ISO range from 200 thru 6400 and “expanded” ISO range from 100 thru 12800, or so. Something about that, begs a question around the generally accepted premise that selecting the lowest ISO gets the best result. I never thought that ISO at lower side would be offered under “expanded range”.

Basically the operative factor is what is called the "native iso" of the sensor. The actual analog voltage output without additional amplification. Generally that is someplace around above ISO 100 and below ISO 200. That is the point were the dynamic range of a recorded image will be the best.

To go a little lower, say for ISO 100 or ISO 50 requires some kind of signal manipulation that cannot possibly be all good! It usually allows whites to clip a little closer to middle gray (less "headroom" for highlights), and may involve just using part of the sensor's capacity that is a bit too non-linear to be called a regular ISO step. But the fact is that for those who want to use a wider aperture in brighter light, it works. It does not provide increased dynamic range, less noise or a better SNR! Just lower sensitivity.

Originally Posted by EinsteinsGhost

One of my conclusions was that, lowest ISO may not be the best bet for best quality (which may also depend on what "quality" entails). So what might be? Should we avoid selecting lowest ISO setting as one normally does with aperture settings on a lens (wide open results are rarely, if at all, sharper than if the lens is stopped down by a stop or two or three)?

Normally, you'd want to avoid it like a plague.

For best noise and dynamic range, it depends on the camera, but whatever is closest to the native ISO value is best. That is probably ISO 200 or just below that.

Originally Posted by EinsteinsGhost

This led me to a web search and discovery that some have claimed (and demonstrated) that selecting ISO in multiples of 160 can yield better results than the nearest “rounded down” ISO. I wonder if this applies universally, and if there is a reason to this? It could also mean that in a Canon, ISO160 (if selectable) would produce better results than ISO100. Could that be the idea behind expanded range also including lower ISO rating as opposed to it being standard?

I suspect you are very slightly confusing two things. One is the native ISO, which is usually closer to 160 than anything else. Values of 140 to 180 are common.

The other is that at least some Canon cameras and I don't know if that is all of them, most or what, or if there are others that do the same thing, are best left at even ISO levels... 200, 400, 800, 1600, etc. The 1/3rd or 1/2 stop levels are performed with the same analog amplification as the next higher step or some such (I don't remember the exact details and it isn't really important). The effect is that at the fractional steps the noise floor is the same as the next full step, and thus the SNR is actually reduced.

It takes a very critical eye to see the difference though. And I don't know that anyone other than Canon does that. Nikon doesn't, but others might.

Originally Posted by EinsteinsGhost

Based on my personal experience with compact (P&S) cameras with tiny sensors, however, is that lesser is greater. But this doesn't seem to apply for cameras with larger sensors.

Any input on the subject?

Well... that is indeed basically true! But recent cameras have thrown a few kinks into the mix. Lower is still better, but in some odd ways!

The D7000, and almost certainly even more so the D800 from Nikon (I don't know about the new Canon cameras, and they might well be the same) have a very different way of dealing with the signal path than say the D2X generation. With a D2X it was relatively simple... The sensor output at full well capacity was 1.0 volt, and that is what was input to the Analog-Digital-Converter for all of the "standard" ISO values. To get ISO 200 the signal was amplified 2x (an output of 0.5 volts from the sensor arrived at the ADC as 1.0 volts). For ISO 400 is was 4x. At the point where Nikon called it an extended range the analog input stayed the same, so if the ISO were doubled the actual signal would be 0.5 volts instead of 1.0, but on the digital side they just doubled the value. So that 0.5 volt signal, instead of producing a 128 pixel value now produces a 255 value (when converted to an 8 bit JPEG). But if the ADC is a 14 bit ADC, it is now working with one less bit, so it is 13 bit data. The next step makes it so that an analog voltage of 0.25 will produce a 255 value, and the ADC is now working as a 12 bit device.

Well, if you have ever seen the results of going to those extended range values on a D2X, you'll recall that they were best describes as horrible. Noisy. The reason is because the "read noise" of the sensor and the analog ISO amplifier was relatively high, and each time the digital value was doubled it dropped the Signal-to-Noise Ratio and the Dynamic Range by the same amount the ISO was increased.

Along comes the D7000 (and the D800 appears to use this even more effectively), and things have changed a little. The "read noise" is so low that most images are photon noise limited (fluctuations in the arrival rate of photons). Because of that reducing the analog signal as is done with digital ISO multiplication does not reduce the SNR or the Dynamic Range any more than analog ISO multiplication! In fact, it appears as if the D800 might not use analog amplification past about ISO 1600! That also means that you can probably set the camera to ISO 1600 and under expose an image by 4 stops, then take another with it set to 25600 to compare to. And then with an editor increase the brightness of the first one to match the second one, and they will be virtually identical! (Technically, that is just astounding!) That might not work out quite as well as it sounds though...

It appears that Nikon is using another trick to generate higher ISO's. Instead of analog gain or digital manipulations (with the loss of relative bit depth), they have an ADC where they can change to reference voltage that samples are compared to. So when they allow a 0.5 volt signal from the sensor to be "full scale", they change the reference voltage on the ADC from 1 to 0.5, and still get a full 14 bit depth digital signal. That is slightly different than converting a lower bit depth RAW data set to a 16-bit depth TIFF and editing that. Hence the two images above may or may not be "exactly the same", but they will be very close. We might assume that Nikon has very carefully optimized their processing for best results, so it will probably be better than digital multiplication with an editor. (These techniques are not unique to Nikon. I assume Canon is doing similar things, and know for sure that some of the medium format cameras do the same thing. Some of them apparently do not use analog ISO amplification at all!)

As they say, we live in interesting times!

[Floyd_Davidson]

Originally Posted by EinsteinsGhost

Based on my personal experience with compact (P&S) cameras with tiny sensors, however, is that lesser is greater. But this doesn't seem to apply for cameras with larger sensors.

Another interesting aspect, which I didn't think to include in the previous post, applies specifically to those who like you wish to use the camera's JPEG engine and want to only do an absolute minimum of post processing.

Typically (at least up until the D7000 and D800 sensors) the Dynamic Range vs ISO curves for most cameras had a flat spot at the high ISO end and a non-linear descent for Dynamic Range as the ISO went up until some point where a bit of a "knee" occurred. From that point it was a straight line downhill, and a 1 stop increase in ISO would result in exactly a 1 stop reduction in Dynamic Range.

The reason for that was because the noise in the analog amplifier and the ADC, which is constant and generated after the gain control to increase ISO, was greater than the noise generated in the sensor and first stage of the ISO amplifier. So at say ISO 200 the noise from the sensor is perhaps less than a quarter of the noise from the sensor. At the ratio, sensor noise barely has any effect. and so changing the ISO gain and raising it a little has no effect. But doubling the ISO amplification to get ISO 400 doubles the sensor noise, and it is then only half what the ADC contributes. At that point is has a small effect. Now, double it again the two noise sources are equal! That has an effect! Double the ISO again and the roles are reversed, with amplified noise from the sensor (which goes up and down in direct proportion the the ISO) being twice that if the ADC (which now has very little effect). From that point on a one stop change in ISO increases noise also by one stop and reduces Dynamic Range by one stop.

What's all the techie gobble de g00k mean to a photographer? It means that changing the ISO between say ISO 200 and 2000 (on a D3S, but every camera is different) results in relatively little change in dynamic range or noise in respect to what can be used in a print. The numbers will be greater than 5 or 6 fstops, and hence if no contrast or brightness changes are made to an out of camera JPEG before it is printed, the slightly greater noise in shadows at ISO 2000 will be below the black level of the print and the image won't look different than one shot at ISO 200. Going above that ISO will reduce the dynamic range down to 6 or 7 stops, and that will show up in a print.

Hence, for most cameras there is some range from about 200 to something from 640 to 2000 where the JPEG straight out of a camera will not change much in quality with ISO over that range. And it is obviously useful to know about what that range is, and try to use if whenever possible.

[EinsteinsGhost]

For best noise and dynamic range, it depends on the camera, but whatever is closest to the native ISO value is best. That is probably ISO 200 or just below that.

That is something I’ve read as well, typically around 160 (and multiples of it). Which begs the question: How about having a higher native ISO? I would assume that would help shift the availability of maximum dynamic range (which falls rapidly with higher ISO) and somewhat improve SNR (reduce noise) at the higher ISO. The shift may not be linear, but it should still help?

The other is that at least some Canon cameras and I don't know if that is all of them, most or what, or if there are others that do the same thing, are best left at even ISO levels... 200, 400, 800, 1600, etc.

Probably out of convenience and the differences may not be perceivable (for example, degradation from ISO160 to ISO200).

Well... that is indeed basically true! But recent cameras have thrown a few kinks into the mix. Lower is still better, but in some odd ways!

With my good old friend (Sony F828) it is definitely true. Although, I see very little difference between ISO64 (the lowest it allows) and ISO100. But it employs CCD chip instead of CMOS, so there might be differences in how CCD versus CMOS deliver results.

The D7000, and almost certainly even more so the D800 from Nikon (I don't know about the new Canon cameras, and they might well be the same) have a very different way of dealing with the signal path than say the D2X generation. With a D2X it was relatively simple... The sensor output at full well capacity was 1.0 volt, and that is what was input to the Analog-Digital-Converter for all of the "standard" ISO values. To get ISO 200 the signal was amplified 2x (an output of 0.5 volts from the sensor arrived at the ADC as 1.0 volts). For ISO 400 is was 4x. At the point where Nikon called it an extended range the analog input stayed the same, so if the ISO were doubled the actual signal would be 0.5 volts instead of 1.0, but on the digital side they just doubled the value. So that 0.5 volt signal, instead of producing a 128 pixel value now produces a 255 value (when converted to an 8 bit JPEG). But if the ADC is a 14 bit ADC, it is now working with one less bit, so it is 13 bit data. The next step makes it so that an analog voltage of 0.25 will produce a 255 value, and the ADC is now working as a 12 bit device.

Thanks for making an effort to get into technical side of the matter, with an example.

Another interesting aspect, which I didn't think to include in the previous post, applies specifically to those who like you wish to use the camera's JPEG engine and want to only do an absolute minimum of post processing.

Not necessarily. I rarely used the RAW mode in my Sony F828 but mostly because of space limitations, the storage issue as well as the camera processing time for RAW. Neither is an issue, with cheap storage and abilities of today’s camera to take RAW+JPEG at full resolution without delay. For most part, yes, I’m a post-processing minimalist now. When I feel there might be a risk involved (being a photographer at a major family event, or trying something different as I just did, turning off long exposure NR in camera and see “play” of various ISO settings for long exposures), I normally go for RAW+JPEG. Generally speaking, though, I would rather not spend too much of my limited free time working in LR or PS.


In most modern day interchangeable lens cameras now, as is true with even my little Sony NEX, noise is pretty much a non-issue at even 1600. And when it does become an issue, post processing can pretty much take care of it (and getting better).

[Nomadicus]

I can do some test shots with not full frame but larger than the 10D-7D types. 1.3 crop VS 1.6. Interesting concept. I know that I liked to print at a resolution that would divide evenly into the printer is using. In other words using any size Epson photo printer using 1440 dpi output I normally sized the image at 240 dpi. 1440/240 = 6. I don't remember where I read something on this but it makes sense that maybe the same will hold true for other non film cameras where you can set the ISO. I've been a habitual go for fine sharp images. A habit formed in my stubborn head since the days of Kodachrome 25.

[Floyd_Davidson]

Originally Posted by Nomadicus

I can do some test shots with not full frame but larger than the 10D-7D types. 1.3 crop VS 1.6. Interesting concept. I know that I liked to print at a resolution that would divide evenly into the printer is using. In other words using any size Epson photo printer using 1440 dpi output I normally sized the image at 240 dpi. 1440/240 = 6. I don't remember where I read something on this but it makes sense that maybe the same will hold true for other non film cameras where you can set the ISO. I've been a habitual go for fine sharp images. A habit formed in my stubborn head since the days of Kodachrome 25.

The image has Pixels Per Inch, though often that is referred to as DPI. The printer has Dots Per Inch, but they are never pixels per inch. Two very different things!

The image has dimensions in pixels only, not dots. The printer prints in dots of ink and each pixel from the image is necessarily made from multiple different colored ink dots.

But the difficult part is that the print driver does not send data to the printer one pixel at a time. The input image is defined with pixel dimensions, but the driver converts an entire line of pixels directly into an entire line of ink droplets, dithering and then encapsulating date into a matrix of dots. The matrix might be 8x8, or 16x16 or some other value. The size of the ink droplets varies according to the resolution of the printer configuration.

With Epson printers the optimum image dimensions for photographs is 360 pixel per inch. If it is anything else (for example 240 PPI), the print driver absolutely converts it in a way that effectively produces a 360 PPI output. It is all done in one step, and the actual data sent to the printer is for the entire line.

The primary reason that 360 PPI is optimal may or may not make any difference to some people though! Ideally an image should have sharpening applied at the size at which it will be displayed. That is because resizing an image changes the effect of the sharpening. It is not possible to look at an image with pixel dimensions of 2880x3600 (which would print at 8x10 inches at 360 PPI) and determine what the real effect of sharpening will be if it is then printed at any size significantly different that 8x10. For example, if the print is actally going to be 16x20 the only way to judge the effect of sharpening is to first resize the image to 5760x7200 and then sharpen it and send it to the printer.

For those who are not super critical though... if close enough is good enough (for example if a program default setting is used to determine how much sharpen to use, rather than a visual inspection of the actual effect), it probably doesn't make any difference at all whether you resize to 360 PPI or let the print driver do effectively the same thing.

But 240 PPI is not optimal for any printer. And while Epson prints at 360 PPI, HP and Canon print at 300 PPI, and various commercial printers more at different values from 200 PPI to 406.4 PPI (Fuji's LightJet 5000 printers).

[Nomadicus]

Originally Posted by Floyd_Davidson

The image has Pixels Per Inch, though often that is referred to as DPI. The printer has Dots Per Inch, but they are [i][b]never[\b][\i] pixels per inch. Two very different things!

The image has dimensions in pixels only, not dots. The printer prints in dots of ink and each pixel from the image is necessarily made from multiple different colored ink dots.

But the difficult part is that the print driver does not send data to the printer one pixel at a time. The input image is defined with pixel dimensions, but the driver converts an entire line of pixels directly into an entire line of ink droplets, dithering and then encapsulating date into a matrix of dots. The matrix might be 8x8, or 16x16 or some other value. The size of the ink droplets varies according to the resolution of the printer configuration.

With Epson printers the optimum image dimensions for photographs is 360 pixel per inch. If it is anything else (for example 240 PPI), the print driver absolutely converts it in a way that effectively produces a 360 PPI output. It is all done in one step, and the actual data sent to the printer is for the entire line.

The primary reason that 360 PPI is optimal may or may not make any difference to some people though! Ideally an image should have sharpening applied at the size at which it will be displayed. That is because resizing an image changes the effect of the sharpening. It is not possible to look at an image with pixel dimensions of 2880x3600 (which would print at 8x10 inches at 360 PPI) and determine what the real effect of sharpening will be if it is then printed at any size significantly different that 8x10. For example, if the print is actally going to be 16x20 the only way to judge the effect of sharpening is to first resize the image to 5760x7200 and then sharpen it and send it to the printer.

For those who are not super critical though... if close enough is good enough (for example if a program default setting is used to determine how much sharpen to use, rather than a visual inspection of the actual effect), it probably doesn't make any difference at all whether you resize to 360 PPI or let the print driver do effectively the same thing.

But 240 PPI is not optimal for any printer. And while Epson prints at 360 PPI, HP and Canon print at 300 PPI, and various commercial printers more at different values from 200 PPI to 406.4 PPI (Fuji's LightJet 5000 printers).

I understand. When using canvas with it's texture you can get away with things that you can't on glossy photo paper. I've seen some very large images done on the Fugi LightJet printers at 200 dots per inch. Most impressive. I have printed many a poster print at 180 dots per inch with success. To know it would take jeweler's glasses or a loupe maybe but the clientele never new the difference. As for optimum ISO I've turned on the expansion feature in my digital to allow ISO 50 to ISO 3200. It's the pits shooting at ISO 3200 in an arena with poor lighting and the images coming out still at 3 stops under when flash and strobes are not an option. From what I have read the expanded range on my camera is not a true ISO as are the settings from ISO 100 -1600.

I mess up when typing info on ppi and dpi often. I know the difference just still mess up. Guilty.

[Floyd_Davidson]

Originally Posted by Nomadicus

I understand. When using canvas with it's texture you can get away with things that you can't on glossy photo paper. I've seen some very large images done on the Fugi LightJet printers at 200 dots per inch. Most impressive. I have printed many a poster print at 180 dots per inch with success.

The older Fuji LightJet printers did print at 200 PPI, while the newer LightJet 5000 prints at either 304.8 PPI or 406.4 PPI.

Regardless, the point is that no matter what your image actually has for pixel dimensions, the printer only prints at one rate and the print driver is necessarily going to effectively resample the image to that rate. Hence if you send a 180 PPI image to a printer, unless the printer is configured for 180 PPI (economy mode with an Epson) the printer driver will change it. (If you actually do print it at 180 PPI, it's going to look pretty horrible!)

To know it would take jeweler's glasses or a loupe maybe but the clientele never new the difference.

They could probably see the difference if you printed it twice, using the optimum way and the easy way. But even the easy way looks great, and it's true nobody will notice that it could be better without seeing them side by side.

As for optimum ISO I've turned on the expansion feature in my digital to allow ISO 50 to ISO 3200. It's the pits shooting at ISO 3200 in an arena with poor lighting and the images coming out still at 3 stops under when flash and strobes are not an option. From what I have read the expanded range on my camera is not a true ISO as are the settings from ISO 100 -1600.

The latest crop of camera bodies, from almost all manufacturers, are changing the way that works! They are calling it "ISOless". What it means is that read noise is so low that there is no signficant difference between hiking ISO with an analog amplifier or doing it digitally.

In practice new cameras will have a very restricted range of "real" ISO values, and everything else will be digital manipulation. For example the Nikon D3 series all have analog ISO values and the HI-0.3 etc values are digital. With the D3 that was up to 6400 and the D3S up to 12800. The next generation (D4 and D800) use analog only up to maybe ISO 1200, and everything above that is digital! The extended values aren't done any differently, just that they are beyond the acceptable limits for image quality.

That means we can probably set the camera to ISO 1200, let it underexpose by 3 stops or even more, and then pull the shadows out using an editor, and get just as good a picture as if the camera's ISO had been set "properly".

I don't know specific details for Canon or Pentax, but my understanding is that they are doing essentially the same.

I mess up when typing info on ppi and dpi often. I know the difference just still mess up. Guilty.

Except that business of dividing 1440/240 was not valid.

[Nomadicus]

Except that business of dividing 1440/240 was not valid.

No arguments from me on that one. That was more homespun logic on my part. I did print literally thousands of 5x7 and 8x10 glossy prints at 360 what ever it is. Tack sharp. Normally taken at ISO 100 with the thinking that the lower ISO number would yield a better print. Cloudy days ISO 200 up to ISO 320.