Friday, 21 January 2011

Moon photography and GH2 ETC

The Panasonic GH2 introduced a new concept: Extended Tele Conversion (ETC). This has been much hyped, and perhaps a bit misunderstood.

Will ETC give you a 2.6 tele conversion for any lens? So the Lumix G 20mm f/1.7 becomes a 52mm f/1.7 lens? The much anticipated portrait prime lens? The short answer is: Sort of, but only for video.

What ETC does, is simply to record the pixels in the centre, rather than the whole sensor area. It is, quite simply, a digital zoom. Digital zooms have been around for ages.

So why is this big news? Because it also works for video. If you record a video at full 1920x1080 resolution with the GH2, it is sampling the pixels from the whole 4976x2800 sensor area, and downscaling it to 1920x1080. In ETC mode, it simply samples the 1920x1080 pixel area in the centre of the sensor, and does no downscaling at all. So you still get full resolution, but with a 2.6 crop factor. The crop factor is calculated by dividing 4976 by 1920. Here's an illustration:

This means that when recording movies at full HD resolution, you can use ETC to convert the Lumix G 45-200mm f/4-5.6 zoom lens into a 117-520mm f/4-5.6 lens. In terms of a traditional 35mm film camera, that is equivalent to 234-1040mm range, since the Four Thirds format has a 2x crop factor itself.

My answer above was "sort of". The reason for that is that digital noise will be somewhat larger in ETC mode, especially with higher ISO values. This is because there is no downscaling. When using the whole image sensor, the camera can use more pixels to compose the video output, to avoid noise. However, when using only the centre pixels, downscaling is not possible. So the noise level is typically higher. I have made a study which shows that there is indeed more noise, and some less sharpness and contrast when using the ETC mode.

Here is a video recording of the full moon, using the Lumix G 45-200mm f/4-5.6 zoom lens at 200mm f/7.1. The camera was put on a tripod, and I used spot autofocus. I used ISO 250, 1/400s exposure. I used the ETC mode.

You can see that the moon moves in the frame. This is simply due to the rotation of the earth.

To see the effect of not using the ETC mode, here is a still image taken with the same image parameters, but without ETC:

Without the digital zoom from the ETC mode, the moon is just a small disk in the centre of the image, even when using a 200mm tele lens.

To capture an image in which the moon makes up a significant part of the frame, you'll need a very long lens. Let's say you want the moon diagonal to take up half of the Four Thirds sensor diagonal. In that case, you'll need a lens which is 600mm. Such a lens is currently not available for the Micro Four Thirds format. Panasonic and Olympus both produce tele zooms that extend to 300mm, but that's still only half way.

I found the exposure parameters by using spot exposure metering off the moon itself. On the other hand, when I used the normal exposure metering mode, I got this image:

In this case, the moon is just an overexposed blob with no features at all. We also see why there is some lack of sharpness in the moon images: There was in fact a cloud cover. But the cloud was too dark to be seen in the image when metering off the moon.

The difference in the exposure is quite huge, around 8 stops:

Spot metering off the moon: f/7.1, ISO 250, 1/400s (13EV)
Center Weighted: f/7.1, ISO 3200, 1/30s (5EV)

One interesting fact is that the moon spot metering exposure is virtually identical to ISO 100, f/16, 1/60. And why is this significant? Because ISO 100, f/16, 1/60 is known as the Sunny 16 rule. Basically, it says that when exposing objects lit directly by the sun, you can use that exposure as a rule of thumb. And the moon is also an object lit by the sun. It is just a bit farther away from us than what we are used to photographing.

So, the Sunny 16 rule also works for the full moon. But you may still want to expose a little bit more. After all, you don't want the moon to be average gray, you want it to be a bit bright. So you could adapt a similar Loony 11 rule.


  1. ETC could produce less sharp picture than downscaling due to demosaicing. Actual resolution would be less than 1920x1080 RGB pixels since each sensor pixel captures only one color channel.

  2. Yes, this is correct.

    We don't know how many of the 16 megapixels the GH2 samples from when recording videos in full HD format. My guess is only relatively few of them. Perhaps around 4MP, since the high speed mode of the camera can record 4MP?

    Downsampling from the full 16MP sensor during video would require too much processing power.

    Anyway, it is clear that video recorded using the full sensor surface has higher quality than the ETC video.

  3. "ETC could produce less sharp picture than downscaling due to demosaicing. Actual resolution would be less than 1920x1080 RGB pixels since each sensor pixel captures only one color channel."

    I think I disagree with both of those statements. De-mosaicing is done on the full sensor images, but is not required for the ETC mode. The ETC mode ("crop mode") has no need for this or for any anti-aliasing, anti-moire calculations because all of the pixels are adjacent. Also, each of the pixels in the 1920x1080 center of the sensor captures all three colors.