Jump to content

Classified no longer - how the Nikon D800's 36MP sensor line-skips for 1080p

Andrew Reid

Recommended Posts

  • Administrators

Falk Lumo has done a great technical analysis of the Nikon D800 and how it obtains 1080p video from the huge 36MP Sony sensor inside.

[b][url=http://www.eoshd.com/content/7788/classified-no-longer-how-the-nikon-d800s-36mp-sensor-line-skips-for-1080p]Read full article[/url][/b]
Link to post
Share on other sites
I was mentioning awhile back Canon should offered 2 versions of the MD3, with a higher price for better video.  As stated above, Canon is having another version in reality with their 4k DSLR version, which is a modified MD3?  Alot of questions remain.  Price?  Actual image quality?  Compression?  Release date? Lot's of competition coming so you snooze you could lose.  I did read a rumour awhile back about a sub C300 and an above C300.  Only a rumour. We'll see.  Answers coming.
Link to post
Share on other sites
Falk is actually very brilliant, he managed to figure out how the old Pentax K-7 generated its video by measuring it in the studio. These results look like [s]Sony[/s] Nikon has used the same technique Canon used for the MKII and (presumably) every other Canon VDSLR up until the 1Dx and MkIII.

One note though, I think [s]Sony[/s] Nikon is binning three pixels of like color on the rows exactly like the MKII did instead of discretely reading every pixel value on every third row; if it didn't bin there would be a very noticeable increase in resolution on the x-axis rather than merely a decrease in moire. If the three colors are binned on-sensor it would also cut readout times to more realistic levels. The internal resulting Bayer pattern would be 2240x1260 which is then downscaled to 1920x1080 (the original MkII had something like an 1872x1053 pattern upscaled to fit a 1080 frame). Note that since both would use new, smaller Bayer patterns their true "resolution" would still not max out 1080p, I think the best they could theoretically achieve is about 70.7% of those numbers. His zone plate test chart I did on an old Canon gives results very similar to what the D800 shows. I believe I also ran his test plate once on a GH13, don't know if I still have the results but it did show the GH13 handling the moire better and had slightly increased resolution in the circles, particularly around when the circles were at increments of 45 degrees (sort of like how the D800 does when zoomed in only without the ugly color moire)

My guess is that Canon is now simply doing the same routine on the MKIII as it did on the MKII but does it to both rows and columns (so now 9 pixels of the same color get binned instead of 3).

The only problem I have with Falk's analysis is his conjecture that zooming in adjusts the line skipping to every 2nd row instead of every 3rd row. I just don't see how that could work, you would end up with a new Bayer pattern that would be nothing but RGGR or GBBG, either the reds or the blues would have to be tossed (there might be a workaround, such as perhaps occasionally sampling the odd R or B from one of the "dead" rows. I'm going to have to think about that, maybe I'll write to him). I did notice that resolution on the x-axis improves much moreso than the resolution on the y-axis when zoomed in. That suggests that the sensor either quits binning the like colors on the x-axis and reads out each individual pixel (it could just bin in pairs I suppose).

Falk had caught Samsung using a different sampling routine on the K-7 when you zoomed in to 8x and 10x, and I think in their description of the K-7's sensor Samsung had in fact described that it had two sampling modes for live view so he's very good at catching these things (it never would do a simple 1:1 readout even when zoomed in to the maximum level). He and DSPographer are probably the most competent sources on the internet on this subject and if he states something in his blog it is highly likely to be either the real deal or at least very close to it.
Link to post
Share on other sites
Awesome, I find information like this devilishly interesting. But when you take that resolution into account that's actually a very tiny amount of surface area of the chip that is capturing data. Like the pixel pitch is near the same as a GH2 isn't it? So even if they captured data in the same way the D800 data would be spread out over an area four times as large meaning the gaps between data points would be four times as large as well.

I'm interested to see how the 1Dx fairs, that's going to be our last full frame stills camera with video support for probably a couple years.  I think the 1dx has the  most interesting compression method because it's the only stills camera that is going to be doing intra frame encoding which is much more efficient.
Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Create New...