Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
Please, support PV!
It allows to keep PV going, with more focus towards AI, but keeping be one of the few truly independent places.
Question about GH2 DR
  • 58 Replies sorted by
  • @magnifico Yes I think the binning does it for the low light and DR. If you look at the dxomark test explanation, they do it toward a fix value of 8 megapixel. The d7000 at its 16 megapixel is about 13.5 going up to 13.9 in the 8 megapixel final reference size. The same is true for Noise performance.

    I wrote a threat about the gh2 being king of low noise dslr (tittle change by vitaliy) with the hack, I thnk it is due to the down-scaling/binning that is narrowing the gap between the gh2 and the larger sensor Canon. In photo mode at 1to1 pixel the gh2 is clearly beaten in low light by the Canon and 5d mark 2 in particular, but in video with the binning and higher bitrate the outcome is different.
  • many thanks to pana and to...VK :-)
  • @magnifice

    You're right about pixel binning improving DR because the virtual pixel size is bigger.

    DXO does some great analysis - but their numbers are absolutes, they don't factor S/N into their DR measurements. DPReview, on the other hand does - their DR measurements use either the lowest pixel value, or the level where low level detail is swamped with noise - whichever comes first. It's kind of a combination of DR and S/N - which translates to usable DR. Less technically correct, but more meaningful (IMO) if you simply want to know general sensor performance. The DXO numbers are more difficult to relate to final image quality for me. Look here:

    http://www.imaging-resource.com/PRODS/D7000/D7000IMATEST.HTM

    Down the page there is a chart that compares cameras. You'll notice that raw DR is nearly always considerably higher than JPEG DR (in the case of the D7000 it's 12 vs 9.7 stops). They do that because cramming too much DR into JPEG usually looks bad - and AVC compression is essentially very similar to JPEG. A few cameras do offer higher range JPEGs - but if you look at the result it doesn't look so good.

    Now, if a camera provides pixel-binned raw - that is a different story. You can always trade off resolution for DR. The reason pixel binned high pixel count cameras can't quite reach the quality of bigger pixel wells (at least theoretically) is because the unused space between pixel wells (as a percentage of total sensor area) is higher with high pixel count sensors. Back light sensors largely mitigate this by essentially having no space between wells, but I don't know who is using those these days.

    I've actually compared the low light and high ISO performance between a D3x and a D700. Specifications aside - the D700 does better under those circumstances (even if you downsample the D3x image to match the D700), requiring lower and less obtrusive NR. I haven't used a D7000, but I can see how performance could be similar if the D7000 sensor uses a higher percentage of silicon real estate for actual pixel wells, etc...

    My original comment was that sensor performance is in large part related to pixel well size (or virtual pixel well size); and that, because of the physics involved, can't be improved much. Improvements have to come in the form of various tricks, like multi-image capture, pixel binning, etc... My other point was that if you map, say 14 stops, into 8-bit space the results won't look very good if any post processing is done. Also, raising ISO always decreases DR - again largely because of how many photons can make it into a pixel well space within a given time. If that weren't true how would you explain the effect of ISO on DR?

    By the way, do you shoot exclusively at base ISO? If not, you'll never see the DR ranges you are talking about. Also, think about this: Many people prefer the look a GH2 gives, for example, at ISO 320 over ISO 160. Why is that? For sure it's not because ISO 320 has a higher DR than ISO160.

    Chris

  • @danyell

    The issue isn't strictly sensor size - its pixel well size, so the tradeoff is resolution vs DR. A 16 megapixel image will have more detail than a 4 megapixel image and less DR given the same sensor size. A 4/3 video camera with a 4 megapixel sensor could achieve 15 stops, APS could achieve maybe around 16.5. I think that's approaching the theoretical limit.

    If you look at the Clark Article (http://www.clarkvision.com/articles/does.pixel.size.matter/), almost exactly half way down you'll see a chart that plots dynamic range vs ISO vs pixel well size (pixel pitch). The dash lines are the theoretical maximums based on a perfect sensor. You'll see that the industry is getting fairly close to theoretically optimal limits.

    a 12MP FF camera has a pixel pitch of about 8.5 uM. If you extrapolate on Clark's chart at ISO 160 that comes comes to a little over 14 stops - assuming the sensor is perfect. A 4/3 12MP camera will theoretically max out at one less stop. A 24MP camera FF camera will have the same spec as a 4/3 12 MP camera. Pixel binning has the same effect as having larger sensor wells (approximately). All this assumes one image capture (no HDR etc...). You can also improve DR by using lower ISO, or by combining multiple captures (same effect as lower ISO). My old Kodak SLR/n has an ISO 6 setting (it combines multiple images actually) that can produce an astronomical DR - better than any other camera I've seen. It's kind of a cheat, though.

    Anyway, the upshot is that these camera sensors are getting close to theoretical perfection. The problem with lots of these specs is that they are not normalized - done at different ISOs, image sizes, etc... The physics of maximum dynamic range is incontrovertibly pixel well (virtual) size and available light (and indirectly ISO) limited. You could overlight (brighter than daylight) and do better than these specs indicate as they assume the brightest object is lit by the sun (at least I hope so) - they actually did this with some older movies.

    Modern sensors typically have a noise level of 3-4 electrons per well - that means that they are able to differentiate light levels higher than that (photons are converted to electrons 1:1) - that's pretty impressive. There's not much room for improvement. Dark areas in poorly lit scenes involve 10's of photons per pixel.

    There are so many different ways of doing these specs. True if it's Tuesday and you own a green umbrella you might get these high DR results. Under normal shooting conditions with a normal camera - nah.

    Chris
  • Chris: going on a method like the one Vitaliy suggested (recording multiple exposures simultaneously) I guess it would be possible to virtually push the boundaries quite a bit into something like a HDR output? It would of course make more sense for manufacturers that use smaller sensors to develop this kind of technology..
  • It still comes down to how many photons hit a well in a given time - so that won't help for low light stuff. It could help with well light stuff, though.

    Often people believe that better DR will improve badly lit scenes - generally it won't with modern cameras. The limiting factor with modern sensors is more the physics of light and sensor well size than anything else. When you are at that low a light level the only thing that will improve low light performance is a bigger sensor - whether you are binning several little pixels or using large ones. The greater the total area that the active sensor wells occupy the better. That, and of course using the lowest shutter speed possible. Lowering output resolution (e.g. going from 1080 to 720) will also give you DR (much like binning, or binning more pixels). I suppose downsizing in post would have a similar effect.

    Chris
  • Here's a pretty good write-up I happened upon. He does a better job of explaining all this than I have - and it's specific to the GH2 to boot! he doesn't address pixel binning - but the rest of his explanations are quite good.

    http://www.laesieworks.com/digicom/panasonic-lumix-gh2-noise.html

    It looks like the best way to get another 1 - 1 1/2 stops of sensitivity would be to go to a Foveon sensor, at least theoretically. I wonder is anybody will do that...

    Chris