Nikon 3100 Announced

[Jim Snow]

10 minute clip length, only 720P at 1920 x 1080. Canon doesn't need to worry about this one.

[Steve Phillipps]

I didn't think any of these vDSLRs had the option to record via HDMI? Even the Canon 5d and Nikon D3s - is that not correct?
Steve

[Dan Keaton]

Dear Steve,

As of this date, most all DSLR's, at least those that I am familiar with, were not designed with recording from the HDMI output in mind.

I expect this to change, as the camera manufacturers, at least some, see the advantages of providing a high-quality image to be output via their HDMI.

This allows an external device, such as our nanoFlash, to be able to record:

1. For a longer period of time.
2. To record at a higher quality. i.e. a higher bit rate or using a differnt codec.
3. To record in different file types (MOV, MXF and others).
4. To reduce heat buildup in the cameras during long recordings.
5. To allow their cameras to be more widely used.


The technical challenge is reducing a high pixel count down to HD pixel counts, such as 1920 x 1080 while peserving image quality. Simple line skipping or pixel skipping does not achive a very high quality image for the HDMI output.

[Steve Phillipps]

But this conversion to 1920x1080 is being done anyway in camera to record to the on-board card yes? So there is an HD compliant signal to put of of the HDMI in playback, should be easy to do during record too?
Steve

[Dan Keaton]

Dear Steve,

You have a very interesting point.

The Canon 5D Mark II does not put out 1920 x 1080 via the HDMI output, at least the image size is not the full 1920 x 1080. But the internal recordings are 1920 x 1080.

The key question is image quality. These cameras were never designed to put out a high-quality image via the HDMI output.

Since there is a demand for high-quality images out of these DSLR's, I believe it will come.

[Steve Phillipps]

I assumed that what came out of the HDMI was what was recorded by the camera after it had been through the codec. And by that rationale, if the image through the HDMI is not good then the image recorded to card will be worse.
Steve

[Dan Keaton]

Dear Steve,

On video cameras, the HD-SDI or HDMI input is almost always before compression.

I do not know the signal path of DSLR's. I would not be surprised if some are prior to compression and others after compession.

[Paul Hatcher]

The D95 in September should sort all this out.

[Duncan Craig]

It's been confirmed that the D90 is being replaced by the D7000.

Duncan.

[Michael Murie]

Originally Posted by Daniel Browning

Well, generally they are the same, but sensitivity isn't really what we care about most anyway: it's noise. And when it comes to noise, sensor size (not pixel size) is by far the most important factor (all else being equal).

But doesn't the size of the photosite (the thing that captures the 'pixel') effect the signal-to-noise ratio, and therefore aren't bigger photosites better? Isn't sensor size only important if it allows you to increase the size of the photosites (ignoring electronics and optical limitations.) That seemed to be what people were arguing a few years back: Enough already with the megapixels

[Bill Koehler]

Originally Posted by Jim Snow

10 minute clip length, only 720P at 1920 x 1080. Canon doesn't need to worry about this one.

I think you meant 1920 x 1080p24, otherwise that is an incredible frame rate!

[Daniel Browning]

Originally Posted by Michael Murie

But doesn't the size of the photosite (the thing that captures the 'pixel') effect the signal-to-noise ratio, and therefore aren't bigger photosites better?

Yes, that's generally true for low light (high ISO), but the effect is much, much smaller than is generally assumed. The reason that most people think it is such a large effect is because they make a lot of mistakes in analyzing image noise, such as confusing the effect of pixel size with the effects of sensor size, or measuring different spatial frequencies as if they were the same ("100% crop").

For example, quadrupling the sensor area has a huge effect on noise, allowing you to increase ISO by two stops for the same noise level (use ISO 1600 instead of ISO 400). But quadrupling the pixel area (without increasing the overall sensor area) only allows you to increase ISO by a little bit for the same noise level, maybe 1/3 stop. (The amount of improvement corresponds to the nonlinearity of the scaling relationship between spatial-frequency-normalized read noise and pixel diameter.)

In some cases, there is no benefit at all. For example, the 5D2 has about 1.3 stops less noise in low light than the 7D (illustrating the benefit of overall size). But if you crop the center portion out of the 5D2 (so that the sensor size is the same as the 7D), the noise level is about the same (slight advantage to the 7D, actually). In other words, even if the 7D had the 5D2's larger pixels, noise would not improve. But that is only because of the specific implementation of those two cameras.

Originally Posted by Michael Murie

That seemed to be what people were arguing a few years back: Enough already with the megapixels

The central point of the article seems to be that pixel count alone is a poor indicator of low light performance. I agree with that.

There are some other points in the article that seem to contradict what I'm saying, but really they don't, it's just a different way of looking at it. For example, as an analogy, one of the points made in the article is that larger pistons have more horsepower than smaller pistons, so if you have a certain number of pistons, then larger pistons give the engine more horsepower. I think that's a backwards way of looking at things. What I'm saying is that larger engines have more horsepower than smaller engines; piston size has an effect too, but the effect is much smaller than engine size.

The article does have a few errors, such as where it says that larger pixels hold more charge and that this improves SNR. Actually it that improves dynamic range (SNR at maximum S), which has no effect on low light performance. For context, the benefit of such would only exist if the larger pixels were part of a larger sensor, therefore it is the sensor size, not pixel size, that causes the ultimate benefit to dynamic range. If the sensor size remained the same, then the total charge for a given level of detail (spatial frequency) would also remain the same. (Hence why smaller pixels on an equal-sized sensor do not generally have less dynamic range, at least in the 2-12 micron range.)

The article also cites an example where an older, large pixel camera has less noise than a newer small pixel one. That is pretty rare, so I would guess the author either made a measurement error or just happened to pick one of the exceptions. Almost any other such comparison would show noise remaining the same or improving (e.g. from 10D -> 20D -> 40D -> 50D -> 7D).

[David Heath]

Originally Posted by Daniel Browning

But quadrupling the pixel area (without increasing the overall sensor area) only allows you to increase ISO by a little bit for the same noise level......

Surely "quadrupling the pixel area without increasing the overall sensor area" must mean roughly decreasing the overall no of pixels on the sensor by a factor of 4?

[Jim Snow]

Originally Posted by Bill Koehler

I think you meant 1920 x 1080p24, otherwise that is an incredible frame rate!

Thanks for catching my mistake. That was a pretty bad keyboard 'spasm', I meant "only 720p if you want 30fps."

[Daniel Browning]

Originally Posted by David Heath

Surely "quadrupling the pixel area without increasing the overall sensor area" must mean roughly decreasing the overall no of pixels on the sensor by a factor of 4?

Yes. So the idea is that a 7D with 4.5 MP would have had slightly better low light performance, but a 7D with 72 MP and double the sensor width would have dramatically improved low light performance: two stops higher ISO for the same noise level.