Removing Slight Brightness Variation From Video

[meeshu]

How do you remove slow brightness variations from a video and make it uniform brightness?

Even "professional" software can be suggested to be utilized as I have access to most advanced/professional software.

In the sample clip enclosed, this is from a video of an old 1929 film (which has brightness variation).

sample_.mp4

[lordsmurf]

That's not an error, not a "brightness variable".

What you refer to is lighting in the scene, at the time of shooting. You can see shadows move in the image. Its not a mere film flicker.

That's a Laurel & Hardy short. That looks to have already been restored, scanned by a high quality film master. In fact, that lighting change probably would not have even been visible on a clip that had actual film flicker damage.

[meeshu]

Yes. I didn't describe the issue well. I meant the low frequency flicker, which I understand is common for very old film.

Attached is a copy of the "original" unprocessed clip I believe(?)

There is a definite low frequency flicker seen.

sample_original.mp4

[Alwyn]

Pretty slick restoration. That flicker is annoying. Once you see it...

Off-topic: It could come back to 24fps (I assume that is original); it's currently at 30fps with a duplicate frame every 5th.

[Sharc]

I think this fluctuations are in the original movie and should be left alone.
But if you prefer to 'butcher' it you may want to play with retinex. Attached a quick attempt. Does this look any better to you?
retinex_.mp4

https://github.com/Asd-g/AviSynth-Retinex/releases/tag/1.0.3

[meeshu]

Thanks for the detailed reply!

It is more subjective as to whether you prefer old film flicker or not. The flicker makes it appear nostalgic as expected on most if not all (very) old films. However, I prefer no flicker as flickering can be distracting from the main scene.

I haven't tried Retinex yet, but I noticed that in your sample, the brightness had increased considerably with some flicker still present. This is not really the best result.

Ideally there should be no change in overall brightness and with no flickering present. I don't know what options, if any, there are in Retinex to change the quality of the processed output video?

[Alwyn]

I doubt the original had that flicker. It's too annoying to be deliberately produced like that.

Out of interest, what show was that from? Another version might be on YT to compare.

@Sharc, could you give your Retinex code, I used the 64 bit version (183kb DLL) but it can't be found-"no function named Retinex..."

[meeshu]

I doubt the original had that flicker. It's too annoying to be deliberately produced like that.

The original, as far as I've seen from several sources, does have flicker. Look at my post #3 where I include a copy of the original - sample_original.mp4

Out of interest, what show was that from? Another version might be on YT to compare.

I'd rather not say other than it is from a Laurel and Hardy short movie. Other versions I've looked at also have flicker.

[timtape]

The apparent changes in brightness are not that simple. Look carefully or even better view it on video scopes. It's only the highlights which oscillate in brightness. The darker details like the dark suit remain the same. So it's effectively changes in contrast but with the dark sections remaining at a constant level.

It's hard to believe this was deliberate at the original shoot, and I'm not sure how they would have managed to achieve that changing contrast effect at the shoot if they had wanted to, or even accidentally. Maybe uneven film development in the tank in chemical processing after the shoot. Is this effect seen in the rest of the movie or just this scene?

[Alwyn]

The original, as far as I've seen from several sources, does have flicker. Look at my post #3 where I include a copy of the original - sample_original.mp4

Fair enough, but that isn't the original, it's a digitised copy of the original cine film. I suspect what's happened is that the flickering versions are copies of the original poor transfer. I thought there might be a better transfer around somewhere.

You can use "ColorYUV(gain_y=-25, off_y=-00)" to change the overall brightness ("gain" raises histogram top more than bottom, "off" raises the whole histo) but there are probably better adjusters out there which the experts will point to.

[Sharc]

The apparent changes in brightness are not that simple. Look carefully or even better view it on video scopes. It's only the highlights which oscillate in brightness....

Then I would suggest to apply a mask based on luma for the deflicker filter.
Something like this (play with the threshold and varY parameters):

luma_mask=clip.mt_binarize(threshold=75,upper=false).converttoYV12(interlaced=false).GaussianBlur(varY=1000.0).grayscale()    # build a mask based on luma
overlay(clip, converttoYV24(interlaced=false).<your deflicker filter here>, mask=luma_mask)

For the deflicker filter maybe try Retinex or AutoLevels or AutoAdjust or SmoothLevels etc......or your own temporal luma smoothing function.

Edit:
Here the luma_mask for demo only. The bright areas will be processed by your deflicker filter/function, the dark areas are left alone.

[Sharc]

@Sharc, could you give your Retinex code, I used the 64 bit version (183kb DLL) but it can't be found-"no function named Retinex..."

MSRCP(fulls=false,fulld=false,lower_thr=0.00,upper_thr=0.0001)

Also see the readme.md in the Retinex downloaded .7z for more info.

Well, maybe it's a futile exercise ..... and someone has a better idea ....