Friday, September 3, 2010

Dynamic Range Revisited

Wow … the amount of bullsh-t posted to the Fuji Talk Forum (FTF) is sometimes staggering … lately, we have been revisiting the topic of dynamic range (DR.) I have been quite critical of one of the forum members for posting statements that defy logic … his HS10 manages to beat his F200EXR for DR, despite the evidence of his own images that he can neither preserve highlights nor shadows.

So I thought it would be interesting to explore the issue again. I covered a lot of info when I wrote my sensor size comparison with respect to DR, so I will try to bring a fresh perspective here.

In that post, I showed how dramatic the difference is in sizes between large and small sensors. They run in clusters as well, the small sensors are all pretty close in size, the large sensors are fairly close in size and as a group are quite a bit larger than the small sensor cluster. This is demonstrated in the diagram I drew back then, which is repeated here for clarity.

sensor_sizes[2]

Click through to see the actual sizes of the two clusters … it’s rather dramatic.

Now, Dynamic Range is heavily affected by sensor size ,,, mainly as the result of photosites’ ability to store a lot of photons (which is called the photon well size.) A deep well allows the camera to collect a lot of photons without blowing out. A shallow well causes the cam to blow out easily, which requires reduced expoosures for preservation of these highlights. This leads to blocking of shadows, where the faster shutter speeds leave no time for the tiny photosites to gather any photons at all, leaving the shadows a mass of black on the image. We’ve all seen this and the effect is really obvious in harsh light …

Now, there is the tendency to speak in extremes … especially on the FTF. Some people assume that when I say that large sensors have much better DR than small sensors I am really saying that small sensors have none and large sensors cannot blow highlights. So they post moronic rebuttals that also defy logic.

The fact is that the DR of any given sensor is affected by its size and its technology. Technology advances as time goes by and you can see an improvement in the new G11/S90 sensor over the technology of the S100fs, despite the size difference (which is actually minute – represented by the outer two boxes on top of the above diagram when greatly enlarged.) This means that the G11/S90 sensor has improved in its recent generation while the S100fs never got the chance to improve. Fuji switched to EXR technology instead and dropped sensor sizes … twice.

What this all amounts to is that DR, and SNR (noise after normalization) and color depth is all relative. It’s on a continuum … there are no absolutes. And shooting different sensor sizes quickly shows you that the bigger the sensor, the better everything is in the image quality world. All three of these measures of an excellent image track very closely to the sensor size.

We are lucky that there is now a web site established by a well know lab (DXO) that demonstrates the native abilities of various sensors and sensor sizes. I have compared three sensor sizes for each of these values in order to show how closely these measurements follow the sensor sizes. This means that light happens to adhere to the laws of physics … which should come as a surprise to no one.

First, let me take one sentence from the DXOMark.com site discussing what constitutes the minimum bar for excellent imagery in all three categories.

An SNR value of 30dB reflects an excellent image quality. Low-Light ISO is then the highest ISO setting for the camera such that the SNR reaches this 30dB value while keeping a good dynamic range of 9 EVs and a color depth of 18bits.

So we have 30dB SNR, 9EVs DR, and 18bits color depth.

The sensors I chose from the site are the 1/2.3” sensor from the Panasonic FZ28, which is the only sensor of that size on the site. The HS10 has the same sized sensor, and so will track pretty closely. The S100fs is compared as it represents the largest sensor used in small sensor cams in recent years – 2/3”. And for reference I chose the Nikon D5000 with its APS-C sensor. This camera is selling these days for around $529 in Canada, a scant $30 more than the S100fsa and HS10 have sold for here. Of course, you still need to spend $300 to get a decent lens … but look at the numbers and you will see why so many are moving to dSLRs.

So … let’s start with SNR.

sensor_sizes_SNR[1]

It is obvious that the DR tracks ISO by sensor size. The difference at base ISO is there and follows the sensor size fairly closely, but where it gets interesting is the ability to raise ISO to maintain shutter speeds in low light and for action. The D5000 is 3 stops ahead of the S100fs, which is 1 stop ahead of the FZ28. In absolute terms, the S100fs remains excellent up to 200 ISO, while the smallest sensor never achieves excellence at any ISO. The D5000 is only 1/6 stops under excellence at 1600 ISO, and that is below the threshold of visibility, which is 1/3stops.

So far we track sensor sizes pretty much perfectly. These ratios 100:200:1600, which is 1:2:5 as measured in stops of light (EV) also track the crop factor almost perfectly – 5:4:1.5 …

Now, to DR … the controversial topic that got a whole thread deleted on the FTF yesterday.

sensor_sizes_DR[1]

Same thing … excellent maintained up to 200ISO for the smallest sensor, 400ISO for the S100fs and 3200ISO for the D5000. Same ratio, no surprises.

And finally, color depth.

sensor_sizes_BITS[1]

Same ratio … 100:200:1600. Again, no surprises at all …

The bottom line:

  • 1/2.3” sensors cannot quite achieve excellent imagery at any ISO … they are limited by poor noise performance.
  • 2/3” sensors can achieve excellent imagery at 200 ISO.
  • APS-C sensors can achieve excellent imagery at 1600 ISO.

What I don’t show here is how far technology goes for full frame (FF) sensors like that in the D700. The D3x has the highest pixel density and so would be expected to perform the worst. And it does … it falls to 1/3 stops below excellence for SNR at 3200 ISO. That’s just visible. Pretty impressive though, for 24mp. The D700 is 1/6stops below excellence at 3200 ISO, and that is not visible. And the D3s is 1/3 stops below excellence at 6400 ISO! Amazing performance …

Further, I don’t show what happens when you develop technology like the EXR sensor where you combine two exposures into one with the shorter one used for highlights. This extends or compresses DR dramatically. One or two stops … which makes the EXR technology able to get approach (but not quite equal) APS-C dSLRs at low ISOs … this hardware trick is not shared by the HS10 or the S100fs, but we know from experience that it works. Skies remain blue and leaves do not burn out in dappled sun with EXR cams … but the HS10 burns with abandon.

Scotty says it best …

Ye canna change the laws of physics!

And for your entertainment, one of my favorite novelty songs of all time … listen for Scotty’s signature phrase …

2 comments:

john.buckhoff said...

Kim

Thanks for the information. Oddly I came to this blog post from a thread on the FTF. Never the less I was happy to see the D5000 used in your comparison. I had no idea the difference was so vast.

Thanks again for the information
JB

Kim Letkeman said...

JB ... I did not realize that I'd posted that link. Maybe someone else did. Anyway, the D5000 remains a superb camera ... IQ as good as my D300 or even better and some neat features. And of course, really excellent IQ right up to 1600 ISO (although the exposure is critical up there.) Glad you liked the article.