Wednesday, December 14, 2011

Sensor Size versus Expected ISO Equivalence – low light shootout – D3s vs D7000 vs GH2 vs V1 vs X10 vs G12 vs F550EXR vs SX230

Ok … so you are wondering what the heck I’m doing comparing cameras in every sensor class? Well, it struck me that people are making wild claims about new cameras like the X10 because they really don’t know what to expect.

We see comparisons popping up that say that the X10 beats the u4/3 sensors hands down. Could that be true? Perhaps. But maybe we should use some relatively well controlled data to make that judgment. And we need to start with documenting what we should expect as sensor size increases. How many stops per increment?

Well, it turns out that the manufacturers are working to a relatively clear plan where size is concerned. The major increments effectively double the sensor size, which gives you an extra stop of light, which should allow you to shoot approximately one stop higher ISO.

All things being equal, and when exactly does that happen?

So I put together a table based on sensor size data from DPReview.com and this is what I got:

What you see confirms what I suggested. That the sensor area more or less doubles for each increment. You can see that quite easily by looking at the row maked Area Ratio (x:1), which shows the doubling quite clearly. The numbers should be: 1, 2, 4, 8, 16, 32. They come very close to that, with the APS-C and FF sensors being a tad off … but just.

So the real increments are: 1/2.3” –> 2/3” –> 1” –> 4/3” –> APS-C –> FF. The other sensors are compromises that slightly increase image quality but are minor increments. They basically give you some of the benefits of the next highest sensor while saving on cost.

Here is what the graph of area ratio looks like:

So you should expect a lot of improvement as sensor size grows. Well, yes and no. Doubling looks great on the graph … but in fact each leap is much larger than the last on a linear scale as shown above, but is still only one stop. The line would be linear if the scale were appropriately logarithmic.

With a graph like this, we should be seeing nearly identical imagery by incrementing ISO as shown in the bottom line of the table. Which means that the Canon SX230 should shoot 100 ISO about as well as the Nikon D3s at 3200 ISO. There are a few surprises though, as really tiny photosites require extra noise reduction that has an effect on what can actually be achieved.


I’ll display the 800px version of each image and you can judge for yourself. Note that the blog shows everything at 540px across … click the images to see the 800px versions.

Once again: The originals of these images are copyright © imaging-resource.com and please go to that site to see the originals. I claim fair use of their copyright for instructional purposes.

My Analysis:

Ignoring the F550EXR sample, we see that the sensor sizes track the ISO curve rather nicely.

But wait … these are small images so there is not much to see. Let’s see if they really track the curve by looking at 100%. Now, we need to normalize the sizes, so I chose the D3s 12mp size this time. A good compromise between the low resolution cams and the high resolution cams.


The curve is close, that’s for sure. Anomalies are the GH2, the F550EXR and to a much lesser extent the SX230.

Specific comments by camera:

  • D3s – despite being shot at 3200 ISO, this is still the best shot.
  • D7000 – slightly cooler white balance and lower saturation, but otherwise at 1600 ISO it is as nice as the D3s shot. A close second place. Big pixels matter.
  • GH2 – Not very good. And this, in my opinion, remains the best jpeg engine in the u4/3 camp. Why is it that it is so clearly uncompetitive? My thoughts have been along the lines of two issues, which together mean that u4/3 is not meeting its potential at all:
    • they insist on packing very high resolution onto smaller sensors (that may not be as modern as the Nikon sensors); and
    • neither have a first tier jpeg engine.
  • V1 – Very, very nice. Definitely tracking the curve perfectly, and giving the u4/3 sensor a bit of a spanking. Nikon has a pretty excellent sensor technology and a top notch jpeg engine. That much is clear.
  • X10 – Good detail with a funky white balance. Tracks the curve.
  • G12 – Nice. Warm white balance to match the Nikons and similar detail to that of the X10.
  • F550EXR – Yuck … what is wrong with the F550EXR? First DPReview refuses to test it because they had two horrid copies, and here the F550EXR is just awful. Again, it looks like massive blur in the top left corner. Tragic. (Note: My own copy of the F550EXR when shot in RAW makes far better images that this one. Most unfortunate that review sites are never able to show this.)
  • SX230 – Not quite on the curve. There is some clumping of the hair caused by excess noise reduction. And that is caused by the tiny sensor trying to get a clean image, even at base ISO. There are limits …

Conclusion

I’m no expert at the math of light and exposure, but it looks pretty clear to me that the curve is pretty much confirmed empirically when examining the low light images from http://imaging-resource.com … please go to that site to see the originals.

So when you see wild claims like “my X10 stomps my (insert name of u4/3 camera here) you can believe it, since we know that those cameras are not currently competitive in jpeg. Possibly not in RAW either.

But when people start talking about stomping the G12. well that’s just not true in a test such as this. The G12 is pretty close, as it should be. And it does better in the white balance department, with Fuji tracking on the cool side, as they tend to.

And when you see nutty comparisons with full frame dSLRs, remember this test … the D3s at 3200 ISO creates a better image than the X10 at 200 ISO … that’s just how it goes when the sensor is that much larger (and the jpeg engine that much better.)

Perhaps you enjoyed the curves I threw you in this article … (boo …)