Saturday, July 17, 2010

Sensor Sizes, Noise and Dynamic Range

Much ado is often made about sensor sizes. With regards to Fuji, we know that the 2/3” sensor in the S100fs is the largest consumer sensor on the market. Or was. Fuji have discontinued that model in favor of the S200EXR, which has the EXR sensor at 1/1.6” size. Is that a lot smaller? Can yo upicture the relative difference in size? Can you picture the actual sizes superimposed on one another?

And then we have the inevitable comparison made between the various sensor sizes. This comparison works on crop factors, the simplified definition of which is the relative size of the diagonal. Which itself more or less defines the required image circle size. Which defines the required glass size. And so on.

Now, we often get these fun debates about relative image quality. The best bridge cam has a 1/1.6” sensor today. How does this really compare in size to the Full Frame sensor, or the APS-C sensor?

Well, I’ve created a rather large chart to illustrate the differences. First, I show the relative difference between the four more popular sized sensors, all used in modern Fuji cams. The sizes are:

  • 2/3” – S100fs
  • 1/1.6” – S200EXR, F200EXR
  • 1/2” – F70EXR, F80EXR
  • 1/2.3” – HS10

sensor_sizes[1]

Below the extremely enlarged relative sizes, I show the actual sizes (sized for the typical LCD monitor resolution of 100ppi) and you can easily see that the actual differences in small sensor sizes is vanishingly small. In fact, so are the sensors themselves. The largest of them still fits pretty much inside your pinky finger nail.

Note: you *must* click the image to see the large version. That gives you a true reading on the actual sizes under discussion. The actual sizes are, of course, relative :-)

Let’s examine this another way, just to put some concrete numbers to the visual impact of these differences. The difference in *area* from the largest sensor in use in what we must call “small sensor cameras” to the smallest sensor in common use today is a factor of 2.04 … the HS10’s 1/2.3” sensor versus the S100fs 2/3” sensor. (Note that the 1/2.5” sensor was popular over the last few years and is even smaller than the HS10 sensor, but the 1/2.3” Sony BiCMOS sensor is sweeping those older sensors off the table for performance reasons.)

These area size differences compared to the S100fs and the to a Full frame sensor are captured in a table:

image

So what does this all mean?

Well, it is fairly obvious that the two most popular Nikon sensor sizes are rather large when compared with even the largest small sensor. In fact, the table shows that the area difference ranges from 15 times to 30 times. So, where the S100fs can capture twice the light that the HS10 can (one doubling of size) … the D700 can capture 5 stops more light than the HS10 and 4 stops more light than the S100fs … making for a truly massive difference in image quality, as defined by noise and dynamic range, between the entire small sensor cluster and the large sensor cluster. And the differences pile up as sensitivity rises.

This is corroborated by comparing cameras on the DXOMark site. There, you can get a direct read on small sensors versus large sensors in RAW mode (i.e. ignoring jpeg engine differences.)

Here is an example … a D700 (FF sensor) against a D90 (APS-C sensor) against a Canon G11 (1/1.6” sensor) against a Panasonic with a 1/2.3” sensor against an S100fs (2/3” sxensor.)

First, noise.

image

So we see that the sensor sizes are tracking very well here. In fact, if you look at crop factors versus stops of noise difference, you can see a strong correlation. The laws of physics are speaking here.

The graph is straight forward. The middle two lines are the 2/3 and 1/1.6 sensors. And the differences are negligible. No reason to get twisted knickers over it *when shooting RAW.* The 1/2.3 sensor is about 1 stop down from that. The APS-C sensor is 2.5 stops above the best small sensor. and the FF sensor is 3.5 stops. Tracks the 4x crop factor nicely.

In plain English now … you are shooting a birthday party for your kids. You want to get a great shot of the blowing face lit by the candles on the birthday cake. This is going to require 1600 or 3200 ISO.

Well … the smaller sensors are well below the cur off for excellent image quality of 30dB. Thus, you will be hammering the image with noise reduction. On the other hand, the D700 is still at that magic level at 3200 ISO. The D90 at 1600 ISO. So if you nail the exposure with either, you get a frameable karge print. With the smaller sensor, you will be lucky to score a decent WEB image.

Now … dynamic range.

image

This one is not quite as predictable. In a straight up shooting war, the two sensor size groups cluster quite tightly. And the clusters themselves are about 2 to 3 stops apart. Tracking a little less closely than you expect from the crop factors.

But … this is only part of the story. When you actually shoot, say, the D300 against the D700 (I do that), you find that the D700 has significantly cleaner shadow detail. Sometimes the difference is astounding. This is reflected in the latitude for improper exposure. Underexpose with the D300 and you will suffer somewhat. Underexpose the D700 and you will not really notice a drop in quality (even though it is there.) And underexpose a small sensor … well, suffer baby.

So these sensor size clusters … the big ones against the little ones … they matter. The image quality is indisputably better when the sensor is bigger, all other things being equal (and they generally are, despite the hopes and dreams of the staunch defenders of the small sensor.) Scotty said it best, when he shouted “Ye canna change the laws of physics!”

P.S. What makes the EXR sensors special is that they will combine adjacent “like colored” pixels to form a sensor of half the resolution, which drops the pixel density drastically. This, in turn, changes the nature of the noise and generally allows for much less noise reduction, with the attendant dramatic improvement in shadow details and low contrast details. The chroma noise also improves, although it is tough to say why. But the F70EXR specifically is almost chroma noise free in EXR modes … and that’s pretty amazing …

14 comments:

reenie said...

Hi Kim, I don t understand the sensor size naming format . I have a fuji S200 and a F70. i was looking at the f70 sensor size which is 1/2 or half inch . but non of the dimentions in metric match half inch (12.7 mm). The length is 6.4 mm and the width is 4.8mm. I calculated the diagonal to be 8 mm . Confused ??? Paul

Kim Letkeman said...

Paul, sadly the industry uses size definitions that don't really have a clear tie to the actual sizes. It comes from the size of Vidicon tubes from the 1950s. The outer dimension of the glass envelope followed standard sizes that are still used to define sensor sizes, the relationship being that the sensor area is roughly 2/3 of the area of the enclosing circle (I think.) Anyway, read all about it here: http://www.dpreview.com/news/0210/02100402sensorsizes.asp

reenie said...

Thanks for the link Kim. I understand it now. They should just measure the actual diagonal n quote that dimension, that seems much simpler lol. Paul .

Kim Letkeman said...

I definitely agree. Goofy archaic conventions like that help no one.

Pierpaolo said...

Excellent article as usual.
I wish you had a m4/3 sensor to the comparison

Kim Letkeman said...

Thanks so much Pierpaolo, I wish I had a 4/3 cam as well :-)

Specifically, I would actually consider replacing my D300 with the GH1 at this point, as the new sensor in that thing matches the Nikon APS-C sensors for noise, which impresses the heck out of me. They are lower in dynamic range, though, which is an issue.

But for now I am waiting to see what Nikon will do for a large sensor compact.

Pierpaolo said...

Hi Kim,

I had been waiting for Nikon to come up with a large sensor compact camera for more than a year until the end of 2009, but when they did not announce any update for their top model "P6000" I got impatient and bought the GF1 + 20mm lens. I could not resist its size and range-finder look.
Since then I have not used any more my D300. I was also able to take some nice HD videos of my little baby (leaving in the dust my SD MiniDV canon camcorder).
I have been wondering of buying a GH1 too, but each time I went to the shop and I picked it up, the plastic/cheap feel of its construction (as opposed to the D300/GF1) made me ask why I should just let go 1,000 Euro for it.
By the way, let's see what the GH2 will bring to the game...

Kim Letkeman said...

Agreed ... the D300 is a very satisfactory backup for now (I actual travel with the D700) so I can be patient. The next few years should really rock and roll in the EVIL cam market ...

drpankajshukla said...

Hi Kim!
I read your informative article on sensor sizes and it's implication on various image aspects !
I was stuck at the point where the I.Q.of pana LX3 /canon S90 is acknowledged [by DPR] as better than that of the fujifilm F200exr--even in it's SN mode [comparing the SN mode with the downsized image of LX3]where effectively the pixel density is much less !so in theory the F200 image should be better --but is not ![according to DPR in the image comparison section of the F200exr]

Am I missing something ?
Have I misunderstood something --?
I will be obliged if u can kindly explain and oblige !

Thanks in advance for the help that may come from u !

Kim Letkeman said...

drpankajshukla: So here is where jpeg engines and sensor layouts get involved ... the EXR layout creates some fairly unpleasant artifacts with edges that don't look pretty at 100%. But the F200 and F70 specifically have very good control over chroma noise, so the images still look nice once processed. The LX3 has one of the better sensors in a compact, but it is not better than the F200, but rather about as good once downsized, for noise. The F200 still has binning on its side, which means that dynamic range is much higher. DPR in general make a lot of comments that don't seem to match their data ... the F200 for example retains more detail in SN mode, but they go on to say that SN mode makes almost no difference. The problem is getting worse, as the F80 review was horrid, with almost no shooting in binned mode. So the bottom line for me is that the EXR cams do a better job with less effort in the day time, they fit in a pocket, and they offer very decent high ISO (and better range on their lenses than either the LX3 or S90.) That makes them the easy choice for me.

drpankajshukla said...

Thank u for your response !
It helped !

drpankajshukla said...

hi Kim!
I will be obliged if u can kindly suggest a denoising approach to high ISO [1600] images from the F200exr and the F70exr,considering the fact that the jpeg output from these cams already have almost none of the chroma noise .

So using the chroma noise slider does not change much !However moving the luminance noise slider even a little does decrease the visible noise but the detail in the image goes down precipitously!

i even tried preserving the edges but the output looks artificial with almost no surface detail !

I understand that some noise is acceptable since it will not show up in prints !

In your experience what has been the best approach ?[u can suggest steps /measures from CS or topaz denoise v4 or both ]

Kim Letkeman said...

drpankajshukla: You didn't mention the editing software you use. I use ACR 6 and CS5 for all my editing. I added Neat Image and Topaz Denoise to CS5 but they are needed less often now. I like ACR6 and yes, I always leave a little noise, which is invisible in all but the largest prints. Running Neat Image or Topaz can be done two ways in CS5 ... on a layer, in which case you adjust opacity to leave the detail and a bit of noise, or followed immediately by the "edit -> fade..." command, which accomplishes the same thing. So do your best with the sliders and then back off the opacity until the fine details return. And remember that the details look worse at 100% than they will downsized.

drpankajshukla said...

I have CS4 and topaz dnoise 4!
Thanks for your response !