Friday, January 7, 2011

Dynamic Range Extension – Fuji S5 Pro versus Nikon D200 and Nikon D7000 – Should Fuji market a dSLR with an EXR sensor in APS-C size?

The S5 Pro is a rather legendary wedding shooter’s camera. The reason, for those who have lived in a cave for the last 5 years or so, is that it performs hardware dynamic range extension, which is to say it is able to capture a much wider range of tones than conventional sensors.

And a massive range of tone is what you get when a bride in a white dress stands next to a groom in a black tuxedo with the sun blazing off of both of them. You get far more range than a typical sensor can capture.

A typical trick to handle this is to shoot RAW and expose for the highlights, thus preserving detail in the dress. Then, during the post processing, the shadows are lifted to restore detail in the tuxedos.

Now, this only works if there is sufficient dynamic range at the sensor to actually capture some photons in the dark sections of the sensor when the bright sections must be cut off so early. This is the primary reason why really tiny sensors have terrible dynamic range. If the highlights are preserved, the shadows are ruined by noise.

Enter the SR sensor on the S5 Pro (and the S2 Pro and S3 Pro that preceded it.) This sensor has extra large pixels using an octagonal shape and packing the pixels closer together. Thus, their noise levels have always been smaller than the classic rectangular arrangement. The SR sensor then added tiny pixels to augment the image as an effective second sensor designed to shut off early and thus capture the tonal range well above the mid-tone.

image

Once blended back together, these two images represented a much wider range of tones than classic sensors could achieve in the same space. The following chart shows two contemporary dSLRs that happen to share the same body. but that use sensors from different manufacturers. The Fuji S5 Pro uses the SR sensor, and the Nikon D2000 uses the CCD sensor. Both APS-C.

image

The difference at base ISO is two full stops, largely in the highlights. The difference at 800 ISO is 4 full stops. In other words, the Fuji capture 4 times as much range as the others at base ISO, and 16 times as much range as the others at 800 ISO. You’d have to be a bit deficient to see this as anything but an old-fashioned curb-stomping.

Wedding shooters jumped at this in droves. And for good reason. This camera could do in jpeg what most others could not do in RAW.

Update: User Rattymouse from the Fuji forum mentions in the comments that he experimented with his S5 Pro and was able to fully recover a RAW image shot at +4EV. Check the images yourself, they are linked in the comments. This means that the S5Pro had the further advantage of exposing for the shadows and still preserving the highlights. A wedding shooter’s dream. The SR sensor was pretty special.

Of course, this experiment would need vetting with a wedding dress in bright sunlight … i.e., if you shift already blown highlights right by any distance, how much detail is retained in the dress? The answer is possibly none … Ratty’s experiment recovered details in the mid tones, which are already several stops down.

For further information on the topic of RAW capture and gamma, check out Bruce Fraser’s brilliant article:

http://www.adobe.com/digitalimag/pdfs/linear_gamma.pdf

Ok …. so what about the EXR sensor, which has now found its way into several cameras? Well, it can split the sensor into two separate exposures just like the SR sensor. This means that it can be stopped early and blended back to achieve largely the same results. In fact, the F300EXR’s DR400 mode improves DR by two stops, which is exactly what the SR sensor did for the S5 Pro. So yes, we can get a curve just like this if Fuji wants to release a new APS-C EXR sensor.

But that would be against older CCD sensors … how would it fare against the modern CMOS sensors in cameras like the D7000 and K5?

Well … perhaps not so much. The S5 Pro died for two reasons. First, because Fuji stopped building and selling it (duh :-) … and second, because affordable full frame cameras became available and narrowed the gap enough to lessen the appeal in light of the S5 Pro’s performance issues, like noise above 1600 ISO and very slow performance of basic functions like shot to shot times.

image

The difference is now about a stop of dynamic range, but the D700 has 14-bit RAW and magnificent shadow preservation, so a good shooter will have no trouble with the classic white / black sunshine scenario. Plus, the camera is blazing fast.

Update: It’s still quite a race if the S5 pro is regularly over exposed and then recovered, because 14-bit technology is not required if the shadows are all shifted slightly right.

But look what happens when you go indoors … now the dynamic range is massively different. And when we factor in noise:

image

The D700 is 2/3 stops cleaner at ISO 12,800 than the S5 Pro at 3200. That makes the ceremony and the reception actually shootable without flash, which makes all the difference in venues that disallow flash. What was not possible was now almost easy.

So the S5 pro’s generation of sensors, despite the wicked technology, was entirely doomed at this point. If wedding shooters move off, there is no one left but perhaps a few zealots who cannot let go. And yes, it’s still a great camera, just like the D70s is a great camera … but we’re talking weddings here, where conditions are as challenging as it gets, and the pressure is surreal.

Now lets take a look at modern APS-C cameras and see what might be possible. Obviously, if a sensor this good exists for Sony, Nikon and Pentax, then Fuji could engineer something competitive as well.

image

image

Well, the latest generation of sensors have made a massive leap in dynamic range and it shows everywhere (remembering that dynamic range uses the noise floor in its calculation.) So what we now see is that an APS-C camera has exceeded the S5 pro at base ISO, and is only one stop apart by 400 ISO. 800 ISO has the lead halved to 2 stops, which is still good … but if only one ISO sees much of an advantage, is that enough? At 1600, the Fuji can no longer use the exposure extension in hardware and the difference swings to 2 stops for the Nikon. Couple that with a substantial 1.5 to 2 stops of noise improvement and the contest is again over.

What this means in the context of this discussion is that Fuji had better use the very latest CMOS technology in their sensors if they choose to go after the large sensor market with EXR. Else, EXR will show too little of an advantage overall to put up with Fuji’s foibles.

Note that the X100 uses a CMOS sensor with no EXR in site. Except of course that they use the EXR processor … whatever that marketing speak might mean :-)

My bottom line is that Fuji has probably lost any chance of doing some magic in the APS-C market with EXR. The X100 does not use it, and the wedding photographers have full frame cameras now that perform amazing feats of magic in low light.

I think Fuji should take EXR and start looking at u4/3 sensors. They belong to the consortium, and there are many excellent lenses out there, they need only build a 550EXR with a bigger sensor … sexy stuff.

No comments: