Core melt Causing Aliasing?

[Joe Marler]

Thanks. The first frames you posted were in a flat or log profile. The exposure had been “de-flattened” and the histogram stretched. This was done by something besides simple tracker. Do you know for certain that if this was exported without simple tracker it would look OK, or would it also have banding? E.g, what if you experimentally added the same blur to the entire frame? This would help decide if it’s the Coremelt tool or simply the combination of color correction and blur step which triggers the banding.

Another test would be leaving the Coremelt tool in but taking out the color correction and exporting in in the original flat exposure.

BTW while FCPX is pretty good at showing in the viewer a representation of the final export for difficult cases like this I always recommend exporting the clip and examining it, often with more than one tool. E.g, VLC and Quicktime.

Simple tracker might be degrading the image by itself, but it’s suspicious that I’ve seen similar things happen without that plugin.

E.g, an underexposed 8-bit image in log format is often on the threshold of banding when color corrected. Performing normally innocuous steps such as video noise reduction, smoothing (or maybe even blur) can precipitate the banding. Those steps are essentially suppressing the normal grain or noise, thereby creating a smoother gradient. That smoother gradient requires more tonal steps to smoothly represent, and with log-format underexposed 8-bit color, you don’t have those steps.

[Joe Marler]

OK that’s interesting. What if you export the compound clip as 4k ProRes, then re-import that, then apply the SimpleTracker plugin?

Unfortunately I’m leaving for a field assignment and won’t be able to follow up on this for a week.

[Ted Cox]

Thanks, Joe.

[Joe Marler] “The first frames you posted were in a flat or log profile. The exposure had been “de-flattened” and the histogram stretched. This was done by something besides simple tracker.”

I’m not sure I understand. When you say the “first frames” I take it you mean the first two screen grabs I posted. One of the viewed while adding my tracked item to the SimpleTracker plugin. In this frame, both images, the before and after are flat. The second is the final image after color correction (using Apple’s Color Board). Is that color correction step what you’re referencing? Or are you saying it looks like something strange was done with the original flat footage?

[Joe Marler] “Do you know for certain that if this was exported without simple tracker it would look OK, or would it also have banding?”

Now we’re on to something! I exported the Compound Clip prior to adding it to SimpleTracker and… it HAS banding. I’m not sure why the viewer shows a higher quality imagine than I get on a final export (I’m sharing as a Master File -> ProRes 422LT in this case). But that’s helpful to know.

So that eliminates CoreMelt, or my misusage of CoreMelt as the source of the problem. I can totally accept that the trouble is the source footage breaking down. You suggested adding some noise to try to dither the banding. Any other suggestions? Any neat tricks to get what I’m seeing in the viewer to actually export?

[Joe Marler]

[Ted Cox] “Now we’re on to something! I exported the Compound Clip prior to adding it to SimpleTracker and… it HAS banding. I’m not sure why the viewer shows a higher quality imagine than I get on a final export (I’m sharing as a Master File -> ProRes 422LT in this case). But that’s helpful to know.

So that eliminates CoreMelt, or my misusage of CoreMelt as the source of the problem. I can totally accept that the trouble is the source footage breaking down. You suggested adding some noise to try to dither the banding. Any other suggestions? Any neat tricks to get what I’m seeing in the viewer to actually export?”

As I suspected it is probably the blur that’s causing the problem when combined with the flat exposure being on the left side of the histogram. Re the viewer, if it’s set for “best performance” that by itself might be adding noise that prevents the banding. If it’s set for “best quality”, that when combined with a scaled display might be preventing it. The key point is the viewer isn’t doing a better job, it’s actually doing a worse job and those imperfections (whether scaling or noise) are by happenstance suppressing the banding.

I think the last time I encountered this I added some noise or film grain to the smooth region and it lessened the problem. I don’t remember which. There is a built-in “add noise” effect and I also have the 3rd-party FilmConvert plugin; I used one of those, I think in a mask. You will have to experiment.

Another possibility is you could export it as 1080p. If you have not cropped the 4k, in theory 4k 8-bit 4:2:0 can be transcoded to 1080p 10-bit 4:4:4. It would be interesting if a 1080p ProRes export looked any better: https://www.provideocoalition.com/can-4k-4-2-0-8-bit-become-1080p-4-4-4-10-bit-does-it-matter/

In general these are common problems when shooting 8-bit color with a flat profile which is not exposed to the right on the histogram.

[Ted Cox]

Thanks so much for the help and clarifications, Joe. This has been really enlightening.

I hope your field assignment goes well.

[Jeremy Garchow]

How did you make the still frame and what format is it?

[Jeremy Garchow]

And by still frame, I mean the still of the woman.

[Ted Cox]

[Jeremy Garchow] “How did you make the still frame and what format is it?”

I used FCPX’s Share “Save Current Frame” on the original 4K clip outputting to TIFF format.

Thanks, everyone for your thoughts and assistance. Joe was quite correct. In the end, it was a function of underexposed 8-bit footage breaking down due to the addition of the blur. Using FCPX’s “Add Noise” function, I was able to add 0.02% blue noise and eliminate the banding. ☺ The noise is visible, but not distracting, and the shot looks so much better!