350XT Review . . . .

FYI:

 

Phil Askey has just put out his 350XT review.

http://www.dpreview.com/reviews/canoneos350d/

 

Quite an intersting read.

 

A few thoughts (my opinions, not Phil's):

 

1) Image processing is a factor. As the basic cameras get better and

better, the post processing will introduce more differences to the

final results than the actual camera. The chromatic problems in the

DPP results were surprising to me.

 

2) Phil finds it hard to see a huge improvement in images between the

6mp and 8mp generation. (ie, if you are not cropping, 8MP and 6MP

produce similar results. I have heard this from others; 8mp is nice,

but not a huge jump)

 

3)Noise of the 350XT is not quite upto the high 20D standards. In

fact, the noise readings (to my eyes) were quite similar to the

300D/10D results. That is a bit disappointing, but not surprising

given the different (presumably cheaper) sensor in the 350XT.

 

4) Dynamic Range: WHY WON'T ANYONE TALK ABOUT DYNAMIC RANGE? ? ? Phil

ignores this key parameter. This is one of my three hot button issues

on dSLR's. (Dynamic Range, AF sensor *size*, ISO in viewfinder)

All DSLRs have essentially the same dynamic range. It's a factor of the number of bits (all are 12 bit) and the noise level (all are pretty similar). You expose for the highlights, then see how much you can dig out of the shadows. The only thing that can expand dynamic range significantly is a different type of sensor (like Fuji use with their SR sensors).

<p>

It's also difficult to define. It depends how deep into the noisy shadows you want to dig.

<p>

I didn't see significant noise differences between the XT and 20D when I tested them. Whatever lab bench tests show, in actual use the difference is negligable.

<p>

See <a href="http://www.photo.net/equipment/canon/digital_rebel_xt/">http://www.photo.net/equipment/canon/digital_rebel_xt/</a>

guys, any idea how long it will be before we see a roughly 1500 dollar dslr with an expanded dynamic range?

The Fuji S3 sells for $2500 now, and I've seen it advertised at $2100 (not sure if that's real or just bait).

 

I'd guess a year or two will see something from Fuji at $1500 (or you'll be able to get a used S3 even if there's nothing new).

thanks bob --what about one where i can use the canon and canon-compatible lenses sitting around here (assuming can't on fuji, right?)?

So Bob A: Sounds like we are ready for 16 bit sensors, no? Wouldn't that be nice?

The fact that RAW files record 12 bits says almost nothing about the dynamic range they capture. 2^12 is 4096, so all it says is that there are 4096 different intensities recorded. The difference between intensities depends on the number of electrons that trigger an increment in the analogue to digital converter, allowing for signal amplification, having subtracted the allowance for the noise floor, relative to the full well capacity of the pixel. Some numbers by way of illustration: if the full well capacity is 42,960 electrons, and readings below 2,000 are treated as noise, then the least significant bit is worth 10 electrons, and the dynamic range is a ratio of 42960/2000, or 21.48, or just 4.4 stops.

 

Another common fallacy about sensor dynamic range is making an exposure so that highlights are just clipping - all the way to the right on the histogram, and then progressively halving the exposure time until the highlight detail vanishes into noise. What this omits is that sensor dark current depends on exposure time, so the noise component from this reduces as the exposures are shortened. The consequence is that noise in the shadows of the correctly exposed image is several stops higher than it is when the highlights disappear with the very short exposure. The method greatly exaggerates the dynamic range available from a single image capture.

 

DPReview are planning to do dynamic range testing - see this post from Phil Askey:

 

http://forums.dpreview.com/forums/readflat.asp?forum=1000&message=12418945

 

It will be interesting to see what their methdology will be once they have completed their beta testing.

Bob, Thanks! This is the first chance I've had to see your review of the 350 XT. Great job. I look forward to seeing all of your reviews as they are always very fair and unbiased.

"... the dynamic range is a ratio of 42960/2000, or 21.48, or just 4.4 stops."

 

This ratio calculates signal to noise ratio, not dynamic range. Although the calculations are similar, they are not identical. Nor is any mention made of weighting, although probably beyond the scope of this discussion.

 

http://clarkvision.com/imagedetail/film.vs.digital.summary1.html

As I said, the bit depth can affect dynamic range if noise is low enough. I doubt that going to 16 bits will help with room temperature camera sensors, but with low noise sensors (cooled sensors) noise can be low enough that the extra bit depth is usable.

 

While it's possible that very low noise sensors may be available in the future, I doubt it. In camera cooling would be difficult and power hungry and there are quantum based limits on how low the noise can be at room temperature. It's not just a matter of building better sensors and electronics.

 

However, even with uncooled sensors, it's possible that going to 16-bits could help images that needed a LOT of processing. I wouldn't be surprised to see DSLRs going to 16-bits before too long. There are quite a few 16 bit digital backs for medium format cameras right now. If nothing else it would be a sales feature that could be hyped up as an advantage!

I would think that stretching the dynamic range will require the use of higher-bit sensors. Although a large dynamic range is usefull for high-contrast scenes, it reduces image-quality for lower-contrast scenes. Reason for this is that in lower-contrast scenes, only part of the available bits are used. You can see that clearly on any DLSR: the histogram of a low-contrast scene will show that the low- and high-bits are not used.

In other words: a sensor optimized for high-contrast scenes will perform worse than a "normal" sensor in all scenes with "normal" contrast. Right?

The point of my post was to debunk the myth that the bit depth of a file represents the dynamic range captured. My hypothetical example showed that 12 bits could be used to subdivide a range of 4.4 stops - perhaps I can make the point by a hypothetical example in the other direction. If we assume the same 42960 electron well capacity for a pixel, and no noise floor subtraction/ignore level, then 12 bits could be used to represent a range of 15.4 stops - more that the bit depth. The ADC count would increment about every 10.5 electrons in this case.

<I>In other words: a sensor optimized for high-contrast scenes will perform worse than a

"normal" sensor in all scenes with "normal" contrast. Right?</i><P>

 

With a fixed bit depth (e.g, 12 bits), I believe that's correct.

I'm not sure what you mean by a "sensor optimized for low/high contrast".

 

Sensors are intrinsically very linear (unlike film). If they get 1000x more light, they generate a signal 1000x larger. 0f course there's a hard maximum limit. When a pixel is "full", it's full. No gentle roll off (like film).

 

In fact it's very hard to make a non-linear sensor. The non-linear output (which is what you see when you view an image) is a function of the RAW conversion software.

 

The bit depth gives you the maximum theoretical dynamic range. 12 bits gives you (in theory) a 12 stop range, though at the darkest level, the entire stop has a range of only 1 bit! The brightest stop has a range from 2040 to 4096.

 

Of course noise might fill up the bottom 32 levels, so you'd only really have 7 usable stops, plus even if the noise was low, levels would be heavily quantized in the lower bits. With a 7 stop range, the intensity levels from 32 to 64 make up the whole "darkest" stop, while the intensity levels from 2048 to 4096 make up the "brightest" stop.

Hi Bob,

 

I don't think the bit-depth of a sensor determines it dynamic range, it only determines in how many steps the daynamic range will be divided. Compare it to computer-monitors: you can set a monitor to 256 colors, or 16 million. With both settings white and black are identical (and so is the dynamic range). It is the number of steps from black to white that makes the big difference between 256 and 16 million colors.