I am currently cleaning up some super-8 videos. A lot of dust and scratches could be removed with several avisynth scripts using RemoveGrain, Removedirt and so on.
However one of the videos contains a lot of static dirt and it looks like all the tools have serious problem with these static spots.
I also tried the neatvideo demo but it is blurring too much and it isn´t able to handle this noise either.
Some frames from the video:
Another approach would be photoshop, the spot healing doesn´t work unless I´d process each frame manually. Apart from the insane amount of work the spot healing tool
is just copying surrounded areas and I have the feeling that could look weird in a video.
So I thought of using an action by manipulating the luminance of the spots, but as you can see this isn´t working too well either.
I wonder if there is program/plugin that is able to handle this type of dust better.
+ Reply to Thread
Results 1 to 8 of 8
A video is worth a thousand pictures (or words). Please upload an untouched video sample - 10 seconds or so - that shows the problem.
Okay, Here we go...
I generated an image of the spots by averaging many frames together, touched it up a bit by hand, then:
LSmashVideoSource("C:\Users\John\Downloads\usedom_10s.mp4") spots = ImageSource("C:\Users\John\Downloads\usedom_spots.png", start=0, end=framecount, fps=framerate).Invert().ConvertToYV12() spots = spots.ColorYUV(off_y=-95) #return(spots.Histogram()) Overlay(last, spots, mode="add") ColorYUV(off_y=-12) BilinearResize(width/2, height/2) RemoveDirtMC(50)
[Attachment 41556 - Click to enlarge]
Wow, that result looks amazing! I´ll check it out on the whole video. Thanks jagabo!
To average frames I used several Merge() command in a row:
Merge(SelectEven(), SelectOdd()) # average of 2 frames Merge(SelectEven(), SelectOdd()) # average of 4 frames Merge(SelectEven(), SelectOdd()) # average of 8 frames Merge(SelectEven(), SelectOdd()) # average of 16 frames Merge(SelectEven(), SelectOdd()) # average of 32 frames etc.
Thanks, I´ll give that a try if my attempt with ImageJ and 2000 frames fails. Up to now it looks good but takes a while of course.