What's Causing The Difference Here?

[Tony Bluntwidth]

Attached are two images. Both were taken with the same settings, and at ISO 100, on the Nikon D7200. One is a normal JPEG (converted from RAW in-camera) with the Flat profile, while the other is an individual frame from a video sequence using the same Flat profile and exposure settings.

Now, I am aware that the Nikon D7100 was notorious for its FPN properties, but why does this also appear to be the case with the Nikon D7200, only while shooting video? All JPEG images, even the previews, have clean shadows and do not exhibit any FPN, but once I shoot a scene with a wide dynamic range in video mode, the FPN shows up in the shadows. It is most visible using the Flat profile, but even the Standard profile exhibits this in extreme cases. I could not reproduce this phenomenon on my Nikon D850.

Does the difference have something to do with the way video is captured? Regardless, correct me if I'm wrong, but I remember reading somewhere that the Nikon D7200 uses a different sensor from the one used in the Nikon D7100, so what could possibly cause the disparity between image and video capture here?


Single frame captured from 1080p video using the Flat profile. Shadow performance is unusual.


JPEG converted from RAW in-camera. Shadows are exceptionally clean in the Flat profile.

 

[Horshack]

On most cameras, video uses a 12-bit or 10-bit sensor readout, compared to 14-bit sensor readout for most raws. Try changing the D7200 to 12-bit raw mode and see if the shadows get closer to what you see for video.

Also try shooting an out-of-camera JPG and see how it compares to video, as that would incorporate some of the same in-camera post-processing settings as video.

 

[John Sheehy]

On most cameras, video uses a 12-bit or 10-bit sensor readout, compared to 14-bit sensor readout for most raws. Try changing the D7200 to 12-bit raw mode and see if the shadows get closer to what you see for video.

Also try shooting an out-of-camera JPG and see how it compares to video, as that would incorporate some of the same in-camera post-processing settings as video.

It could be 10-bit readout with a camera that old. That was possibly the only way to take 30+ fps back then without extreme line-skipping.

Low bit depth and/or fast readout tends to bring out this kind of horizontal banding noise more.

If it is truly FPN noise, then it will sit there, statically in the video, the same in every frame. If it is random every frame, then it tends to only show when you pause the video or grab a frame.

 

[Tony Bluntwidth]

On most cameras, video uses a 12-bit or 10-bit sensor readout, compared to 14-bit sensor readout for most raws. Try changing the D7200 to 12-bit raw mode and see if the shadows get closer to what you see for video.

I tried this, and the shadows didn't look noticeably different when compared to 14-bit capture.

Also try shooting an out-of-camera JPG and see how it compares to video, as that would incorporate some of the same in-camera post-processing settings as video.

This was what I tried first, but I forgot to mention it in my description. The only difference between a regular JPEG preview, and a JPEG derived from a RAW file in-camera, was extra compression artifacts in the shadows of the former. Then again, I have the camera set to Size Priority with JPEG capture (as I only shoot RAW).

 

[Tony Bluntwidth]

On most cameras, video uses a 12-bit or 10-bit sensor readout, compared to 14-bit sensor readout for most raws. Try changing the D7200 to 12-bit raw mode and see if the shadows get closer to what you see for video.

Also try shooting an out-of-camera JPG and see how it compares to video, as that would incorporate some of the same in-camera post-processing settings as video.

It could be 10-bit readout with a camera that old. That was possibly the only way to take 30+ fps back then without extreme line-skipping.

Interesting... I know of a few 10-bit Panasonic Lumix cameras that don't exhibit the same noise in the shadows.

Low bit depth and/or fast readout tends to bring out this kind of horizontal banding noise more.

I was shooting at 24fps, but now that you mention it, I think that the Nikon D7200 may be using a bit depth lower than 10-bit. I always wondered why the Nikon D5500 I used to own, which had a Sony sensor, could shoot at 60fps in 1080p using the entire sensor area, but the Nikon D7200 can only do so at a 1.3x crop, despite being released a year later...

If it is truly FPN noise, then it will sit there, statically in the video, the same in every frame. If it is random every frame, then it tends to only show when you pause the video or grab a frame.

Knowing this, I don't think it constitutes FPN because the banding patterns keep moving, in a fashion similar to what uncorrected flicker banding would look like. I initially even thought that the banding was caused by flicker, but it occurred in sunlight as well.

 

[Horshack]

Knowing this, I don't think it constitutes FPN because the banding patterns keep moving, in a fashion similar to what uncorrected flicker banding would look like. I initially even thought that the banding was caused by flicker, but it occurred in sunlight as well.

I had another look at your video grab. Those might just be rolling shutter bands - I didn't think so at first since they only appeared to be in the shadows but that may be due to how these old DSLRs did image processing on video. To see for sure try the same experiment with non-cycling light, ie natural light.

Here's my GitHub project with rolling shutter info/explanations:

https://github.com/horshack-dpreview/RollingShutter

 

[Tony Bluntwidth]

Knowing this, I don't think it constitutes FPN because the banding patterns keep moving, in a fashion similar to what uncorrected flicker banding would look like. I initially even thought that the banding was caused by flicker, but it occurred in sunlight as well.

I had another look at your video grab. Those might just be rolling shutter bands - I didn't think so at first since they only appeared to be in the shadows but that may be due to how these old DSLRs did image processing on video. To see for sure try the same experiment with non-cycling light, ie natural light.

I was able to get the banding effect to show up in a blue-hour sky, which is why I added that it occurred in sunlight in the earlier quote. I have also gotten it to show up in the shadows of my interior, when exposing for outdoor window lighting during the day and all artificial lighting is turned off.

I had initially thought that the banding was related to artificial light, but it happens in the shadows regardless of light source. I have also observed that exposing the shadows properly, regardless of ISO used, completely resolves the problem, which leads me to think that it might be a bit-depth (and consequently, dynamic range) issue.

Here's my GitHub project with rolling shutter info/explanations:

https://github.com/horshack-dpreview/RollingShutter

 

[John Sheehy]

On most cameras, video uses a 12-bit or 10-bit sensor readout, compared to 14-bit sensor readout for most raws. Try changing the D7200 to 12-bit raw mode and see if the shadows get closer to what you see for video.

Also try shooting an out-of-camera JPG and see how it compares to video, as that would incorporate some of the same in-camera post-processing settings as video.

It could be 10-bit readout with a camera that old. That was possibly the only way to take 30+ fps back then without extreme line-skipping.

Interesting... I know of a few 10-bit Panasonic Lumix cameras that don't exhibit the same noise in the shadows.

Well, I didn't say that low bit depth has an absolute affect of banding (one-dimensional read noise). The ratio of visible banded noise to random noise varies with camera models. On any given camera, we tend to see more post-gain read noise with the faster readouts and/or shallower bit depths. It's the same way on my Canon R5; fine banding noise in both 12 and 14 bit readout modes, but stronger in 12 bits. The Canon R7 I bought a couple years after the R5 has no visible banding noise, even in 12-bit mode, although I haven't investigated whether there are any video modes that use less than 12 bits.