Thursday, July 23, 2009

Nikon D40 versus Fuji S6000 -- Battle of the Old-Timers

A friend (Ron from the Fuji Talk Forum) has been chatting with me over email for months now on various topics regarding dSLRs and such. He has recently stepped up his search for a new camera for his wife and we've been chatting about several important points regarding image quality and how these points might compare between a small sensor camera like the Fuji S100fs or the Fuji S6000 and a larger sensor camera like the Nikon D40.

These cameras came up because he is considering the S100fs, the current class leader for bridge cameras (high-featured dSLR-like) and he owns the S6000 and the D40.

So since we had been discussion dynamic range (DR) and noise, this is where I'll concentrate this comparison.

We'll use my favorite measurement site for this, as they go to great lengths to equalize all variables and give real comparisons -- I'm talking about, from DXO Labs in France.

They summarize their scales nicely in this paragraph:
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 9EVs and a color depth of 18bits. As cameras improve, the Low-Light ISO will continuously increase, making this scale open.
Let's say then, that you want to shoot in the late evening and you need higher ISO to retain good shutter speeds. But there are a lot of lights around and you want to retain all details, so you insist on at least 9 stops of dynamic range and you set that as your limit for ISO. Or you want to shoot birds late in the evening against the sun. Or you want to capture action in the late evening.

Obviously you would like 30DB of SNR and 9EV of DR, since these values guarantee you very nice images.

So how close can you get with these three cams?

Here is a normalized comparison of dynamic range at the sensor level (i.e. *real* dynamic range as determined by the saturation point of the sensor.)

So you can see here that the S100fs hits the 9EV mark at 400 ISO, while the D40 is above 9EV at 800 ISO and the D5000 is above 9EV at a remarkable 3200 ISO. Wow ...

The SNR graph is next ... does that limit us further?

The S100fs hits the 30DB mark at 200 ISO, which is our new ISO limit. The D40 drops just below (29DB, 1/3 stop) at 800, so we can keep that limit. And the D5000 drops to 29.4DB at 1600 ISO, so that is our new limit for that cam.

For really contrasty scenes, we can obviously push the D5000 to 3200 and the S100fs to 400 ISO with some noise penalty.

So the older dSLR with just over half the resolution remains 2 stops better in noise and 1 stop better in DR while the S5000 is 3 stops better in both categories. Nikon's new CMOS sensor technology kicks some serious butt! And of course the laws of physics make their mark in a big way.
Note that these are normalized graphs. I.e. they are directly measurable against each other as the same numbers should look about the same in prints at any chosen size.
So what does this all mean? Well, shooting weddings with the S100fs, as one example, is an extremely dangerous game. It simply does not have the chops. I want to say that you'd have to be nuts, but there are some decent photographers on the FTF trying it. Still, I think they will graduate upward soon enough.

It means that the newer CMOS technology in Nikon's cameras is simply amazing ... DR that remains excellent to 3200 ISO ... WOW!

It means, in the end, that small sensor cams are good, but limited. They have to be, obviously, because they have really tiny photosites. But try and jam that concept into some of the bricks, er, heads on the FTF.

But wait ... there is more to the story.

Detail retention in low contrast detail like hair, fur, feathers and distant foliage plays a big role in image quality for me. DPReview measure this, although not in a way that can be compared between cameras. But they are very interested in how rising ISO hurts this type of detail. The S100fs, for example, is not very good here. It has what I have dubbed "sledge-hammer" noise reduction even at low ISO.

But we are confirmed RAW shooters on this blog, so that's what I am concerned about. Of course, with the S6000 you must be a patient person to shoot RAW, as it shoots RAW pretty slowly. The S100fs is quite fast handling RAW, although I detected annoying pauses here and there. And there is the ridiculous 23MB RAW file sizes from the S100fs -- enough room to store 13mp of 14-bit data, so why does Fuji need that much to store 11mp? More importantly, why do they not use compression? Everyone else offers it.

Anyway, Ron was kind enough to send me sample images shot on a difficult day with a subject a fair distance away. The subject is a hairy dog doll with a black head and white body. This is the perfect test ...

He shot both images from the same spot with his D40 and his S6000. Now, we know how sharp the S6000 is supposed to be, and he was using the 70-300VR from Nikon, so the optics should be a match.

To be clear, Ron has had a lot of trouble believing that the D40 was all that clean and that it could compete with a lens that is pretty renowned for sharpness. But, of course, there are other important issues ... subject isolation, which helps with the impression of sharpness, and photosite size, which helps you retain maximum per-pixel details.

The dog was sitting in shadow during the test, so that aggravates all the issues, as the background gets over exposed and the dog gets slightly underexposed. Kind of like wildlife and wedding shooting combined. A very good test.

So here are the two full frame images. S6000 followed by D40. I ran Topaz Denoise on them with identical settings to clean out grain and emphasize the details. Topaz handles that task amazingly well with fine detail.

The first thing you notice is that it is much easier to separate the dog's head from the branch in behind with the D40 because of the subject isolation advantage. The branch is well blurred only a few feet behind the dog, and so are the leaves. This helps quite a bit when there is overexposure of the sky, but it helps in general to separate the background away. It obviously reduces distractions as well.
Note that this is the source of the dSLR look that really separates the professional images from the snap shots. People react viscerally to subject isolation done well, and that's one reason why dSLRs are selling so well.
The second thing to notice is that the Nikon image looks a little "hairier" ... which implies that we are seeing a hint of extra detail even at this small size. Click through the images to see the bigger versions.

Now, if you want to crop away the junk and get a doggy portrait (which is *extremely* common when shooting small birds in trees for example), you get a little different look.

Now the difference in detail retention is unmistakable. The D40 has retained massive detail in the fur. Here is it almost too much, but now imagine dark feathers with very subtle veins. If the Fuji had trouble holding detail in white hair, what will it do with darker feathers with subtle veins?

The answer is smooth them out ... something we see every day in shots on the Fuji Talk Forum (FTF.) Some people have managed very nice shots ... one fellow uses flash for every shot for example ... but we usually problems with blow outs all over even the best images ...

No free lunch. Bigger photosites invariably have better dynamic range and lower noise, and even today the older dSLRs hold there own against the latest and largest small sensors.

My conclusion is as it has always been ... you can slice and dice the data any way you like ... but in the end the size of the sensor has a huge effect on the final result. Some choose to ignore this blindingly obvious fact to the utter destruction of there credibility ... those of you who read the FTF regularly know what I mean ...

I am not crusading against the small sensor ... I own the Fuji F11 and the Canon G10, my youngest has had the Fuji F10 for two years, and my eldest took the Fuji Z10 to Europe this past year. I also own the Nikon 990. I am swimming in small sensors.

But when I want to capture the best possible images, I shoot the big dogs ... I don't fool around with small sensors when the images *really* matter to me.

Your mileage may vary ... but if it does, then you are probably delusional :-)

No comments: