Luma De-Flicker Options

[Eluthingol]

Greetings all - Could anyone offer software suggestions on how to tackle evening out the flicker (if that is the proper term) shown in the attached sample? If "flicker" is not the proper term, I apologize, but would most appreciate clarification and a point in the right direction.

Thanks so much!!

Joe

[johnmeyer]

AVISynth Deflicker() function, used either alone or in conjunction with MDegrain2, with DCT=1. You can use really light degraining settings since you probably don't need degraining, but the DCT=1 setting does a really nice job.

I only had time to use Deflicker() by itself and it removed about 80% of the flicker.

[Eluthingol]

Thanks, John...I will play around with it.
--Joe

[Selur]

Since there is no deflicker for Vapoursynth and I'm doing some filter chain testing atm. and I used that source, here's what ReduceFlicker (script-1,script-2), sharpening and denoising does,...

Cu Selur

[Selur]

Another approach using:
0. KillerSpot (Spot removal) + DetailSharpen (to counter some of the later blurring)
1. HQDN3D for the flickering (with a motion mask)
2. mClean for removing the ghosing artifacts created by HQDN3D
3. SMDegrain + DCT for stabilization in smaller regions.(temporalfix would probably be better)
(0-2 are the important steps)

Cu Selur

[johnmeyer]

Here's the result using the defaults for Deflicker() and also doing denoising with DCT=1, just as I recommended above:

Deflicked

I didn't bother to tune the Deflicker() window, but might have been able to get the last of the variations if I used a bigger window or, perhaps, stacked two Deflicker() calls.

[edit]You'll see a lot of other changes, including slight cropping, because the easiest way for me to do this was to run it through my film restoration script. I didn't tune any of those parameters, so you'll see some artifacts from too-aggressive cleaning. The main thing to focus on is the (almost) lack of flicker.

[Alwyn]

John, that Deflicker works well.

[Alwyn]

And it came out OK using Vapourbox's Deflicker, Spotless and SMDegrain filter...

[VHS-Digital-Buddy]

Hi folks,
Anyone want to take a stab at this ? It's a segment of a full tape that I captured this morning.
Don't worry about the quality of the capture. I just want the flicker gone.

https://drive.google.com/drive/folders/1i42uxV3NoCkgqaiKTuU8dkiE17hNWtjg


I tried the suggestions in the thread, but either I'm doing them wrong, or something else is not working.

Any help will be much appreciated.

[Selur]

I attached what Spotless+Small_Deflicker+VHSClean does,..

[johnmeyer]

The football game video has different problems than the OP's film clip. The football clip is interlaced, and the flashes only happen on one field. A better solution for fixing this would be to bob the result, detect the flash (pretty easy to do), replace with a motion-estimated from the adjacent fields, and then weave the result back together.

[Sharc]

..... and the flashes only happen on one field.....

Yes. The flashes can happen with an even or an odd field. Out of curiosity: Any idea as to what may have caused this? "Normal" camera flashlights rather not, I think....

[johnmeyer]

Out of curiosity: Any idea as to what may have caused this? "Normal" camera flashlights rather not, I think....

It looks to me like a VCR problem. Other people may have a better explanation.

[Alwyn]

@Selur, that looks good, comparatively speaking.

[jagabo]

Using a script that shows the difference between subsequent even fields:

SeparateFields()
SelectEven()
Subtract(last, Loop(2,0,0))

[Attachment 93771 - Click to enlarge]

You can see that there is a bend in the waveform monitor in the lower 60 percent of the field. This is the flicker over that lower part of the frame. There is also some lesser flickering over the top 40 percent of the fields but I'm not going to address that here.

You can also see there is horizontal time base shift (the vertical black and white bars at the left and right of the frame) of about 2 pixels each time the flicker occurs. These might be a good runtime test for substituting a "fixed" field each time one of these bright fields is detected. Or you might use the average brightness of a patch from the subtracted images as the test.

In the attached video I've replaced the "bad" fields with fields interpolated from the field before and the field after (just because I have a filter that does this) using ConditionalFilter(). So the flickering is replaced by a vertical bounce. But it shows that the brightness flickering is nearly gone. Ie, this is merely a proof of concept. I won't have time to address this further until later in the week.

[johnmeyer]

I was going to post a script here to fix this, but the detection code I've used successfully in the past for "flash frames" (like what you get from a photo flash), wasn't sensitive or reliable enough. Here is a script I posted over doom9.org which is designed for detecting and then replacing "bad" frames:

Finding individual "bad" frames in video; save frame number; or repair

I thought I could use that old code, simply making adjustments to perform the fix on fields instead of frames, but I got hung up when my detection code didn't work well enough. If I get a chance, I may go back and look at that detection logic again. The flashes in the sample clip should be easy to detect if I just think about it some more.

[Sharc]

Using a script that shows the difference between subsequent even fields:

SeparateFields()
SelectEven()
Subtract(last, Loop(2,0,0))

Cool analysis method

[Selur]

script: https://pastebin.com/qtGMck0Z see
main idea: use TFM analysis results as basis for when to modify the clip

Cu Selur

PPs.: it has problems with scene changes, but the general idea seems feasible.

PPPs.: turns out using the average luma (comparing it to neighbors) seems to be the more stable solution