How do I remove these vertical lines?

[coolgit]

I have 3 bluish vertical lines that exist in certain scenes. No idea why.

How do I remove it. Currently the bluish is the only colour, the rest of the scenes are blackish, brownish, greyish, therefore i was thinking of a way to remove blue and somewhat blend in the lines with nearest pixel from both sides.

I have a vid sample, 2 seconds.

cheers

[johnmeyer]

I think it will be tough to remove these without creating a blur artifact that will be more obnoxious than the blue line. However, if you want to try, since the lines appear to be perfectly static, you could try the Delogo filter. Here is a tutorial I created, showing how to use it to remove a small sensor dust spot:

https://www.youtube.com/watch?v=Z12TutFSg8c

[coolgit]

That didn't work too well in certain lighter areas, over face, but ok in darker areas. Clone tools in photoshop does better but 3000 frames of cloning is not to my liking.,

[coolgit]

The latest is i have been making headway but 3/4 pixels width of white lines remains.

What i am looking for, dunno if it can work, it a simple replace bad pixel type plugin. The way i see it would work like this.

Instruct start hor pixel(say 300) and end hor pixel (305) and full vert length - like masking and only work in this area and nowhere else.
Pixel 300 to 305 are considered bad.
Lets focus on 5 main colours. BWRGb. | = start and end mask boundary.


Small example

Good bad good
pixels pixels pixels
BBBB|WbbbWW|BBBB
BBBB|WWbbbW|BBBB
GGGG|WWbbbW|RRRR
GGGG|WbbbWW|RRRR

would become by using previous good hor pixel(299) and hor pixel(306) on both side alternately to overwrite the bad pixels

BBBB|BBBBBB|BBBB
BBBB|BBBBBB|BBBB
GGGG|GGGRRR|RRRR
GGGG|GGGRRR|RRRR

The above looks simple enough to me. No blurring, no band lines, no random guessing based on fancy algorithms, etc.

How hard can it be?

[jagabo]

Dekafka() in AviSynth does something like what you're asking for:

LWLibavVideoSource("SH_2_secs.avi") 
ConvertToYV24() # must be YV24 or RGB to crop on odd pixel boundaries

dekafka(last, 363, 2, 4, height-4, 256) # X, Y, W, H, strength (0=none, 256=full)
dekafka(last, 578, 2, 4, height-4, 256)
dekafka(last, 792, 2, 4, height-4, 256)

Since dekafka() uses pixels outside the block to infill -- it cannot fill all the way to any edge of the frame. Hence I started at Y=2 and ended at Y=height-4. Infilling like this will be very noticeable at times.

[coolgit]

jagabo

How do i get it to work. There's no plugin.

All i get is "there no function named dekafka"

What does last mean?
363 means start?
4 means number of pixel to replace?
where does dekafka get the pixel from to infill?

[jagabo]

How do i get it to work. There's no plugin.

Copy/paste from http://avisynth.nl/index.php/DeKafka

What does last mean?

The video you want to process.

363 means start?

The x position of top right of the area you want to process.

4 means number of pixel to replace?

The width of the area you want to process.

where does dekafka get the pixel from to infill?

The area immediately outside the area you specify.

[coolgit]

where does dekafka get the pixel from to infill?

The area immediately outside the area you specify.

How do i specify the pixel outside the area?

OK this is what i did.
Open vdub2
tools -> script editor

paste

function dekafka(clip clip, int Xstart, int Ystart, int X, int Y, int Amount)
{
  ytop = Ystart
  ybot = Ystart + Y
  xleft = Xstart
  xright = Xstart + X
  topline = clip.Crop(Xstart, ytop-2, X, 2)
  bottomline = clip.Crop(Xstart, ybot, X, 2)
  leftline = clip.Crop(xleft-2, ytop, 2, Y)
  rightline = clip.Crop(xright, ytop, 2, Y)
  logosrc_hor = StackVertical(topline, bottomline).Blur(0, 1.58).BilinearResize(X, Y)
  logosrc_ver = StackHorizontal(leftline, rightline).Blur(1.58, 0).BilinearResize(X, Y)
  Amount2 = (Y>=2*X) ? 255 : 128*Y/X
  # Amount2 is small if X >> Y => logoscr_hor is dominant
  logosrc = Layer(logosrc_hor, logosrc_ver, "add", Amount2)
  clip = clip.Layer(logosrc, "add", Amount, Xstart, Ystart)
  return clip
}

FFVideoSource("E:\currser\BABYLON5\new\SH_2_secs.avi") 
ConvertToYV24() # must be YV24 or RGB to crop on odd pixel boundaries

dekafka(last, 361, 2, 6, height-4, 256) # X, Y, W, H, strength (0=none, 256=full)
dekafka(last, 576, 2, 6, height-4, 256)
dekafka(last, 794, 2, 6, height-4, 256)

save and open avi
run analysis pass

nothing happen.

[jagabo]

You changed the starting Y position from 2 to 4. So you need to reduce the height of the block. Change height-4 to height-6.

[coolgit]

I had realised that and turned it back to 2.

Still nothing happened.

In logical order what exactly does this do?

[jagabo]

What do you mean by "nothing happen"? Why are you using run analysis pass? That just runs through the entire script as fast as possible (only useful for 2-pass filters and benchmarking). If you just scrub through the video in VirtualDub you should see the vertical lines are cleaned up. You may need to fine tune the parameters until you're happy with the result. Then save the result.

[coolgit]

Previously when i mod, the original viewer (left vdub) stays the same and the processed viewer changes and therefore i can see the differences. For some reason your code changes both viewers and therefore both looked the same, despite the fact it worked, but gave me the impression that nothing was working.

However this doesn't remotely take outside pixel to replace 'bad' pixel. It just blur the 'bad' pixels thus making it worse. my post no. 4 did say no blurring.

A simple question. Does avisynth have the code ability to copy a specific pixel and paste it in a specific place to replace another pixel? No blurring, no band lines, no random guessing based on fancy algorithms, etc.

[coolgit]

Jagabo.

The file in the above examples was the source file. However i used photoshop to remove the bluish colour with some other fixes and was left with thin white lines, 2/3 pixel width. Last week i tried descratch on it and while it did find the thin white lines it also went rampant in other areas of the frames, when debug mode was on. So i scrapped that idea.

I was thinking is it possible to use descratch in your example by replace Blur(0, 1.58) etc. Surely this would prevent descratch from touching the rest of the frames.

[jagabo]

Previously when i mod, the original viewer (left vdub) stays the same and the processed viewer changes and therefore i can see the differences. For some reason your code changes both viewers and therefore both looked the same, despite the fact it worked, but gave me the impression that nothing was working.

I guess you figured this out but... The output pane in VirtualDub reflects the changes made by VirtualDub's filtering system. When you use AviSynth all the changes happen before VirtualDub gets the video. So the input and output panes in VirtualDub are identical. You can use VirtualDub's filtering in addition to AviSynth.

However this doesn't remotely take outside pixel to replace 'bad' pixel. It just blur the 'bad' pixels thus making it worse. my post no. 4 did say no blurring.

Copying the edge pixels inward usually looks worse than the bilinear fill that dekafka is performing. I made a filter called InCopy444() that does a simple copy like you asked for:

function InCopy444(clip v, int X, int Y, int W, int H, float alpha)
{
    hzfill = StackHorizontal(v.Crop(X-1, Y, 1, H), v.Crop(X+W, Y, 1, H)).PointResize(W, H)
    vtfill = StackVertical(v.Crop(X, Y-1, W, 1), v.Crop(X, Y+H, W, 1)).PointResize(W, H)
    H > W ? Overlay(v, hzfill, X, Y, opacity=alpha) : Overlay(v, vtfill, X, Y, opacity=alpha)
}    

FFVideoSource("SH_2_secs.avi") 
ConvertToYV24() # must be YV24 or RGB to crop on odd pixel boundaries

InCopy444(last, 362, 2, 6, height-4, 1.0) # X, Y, W, H, strength (0=none, 1.0=full)
InCopy444(last, 577, 2, 6, height-4, 1.0)
InCopy444(last, 791, 2, 6, height-4, 1.0)

[poisondeathray]

*In general you want to use the original DVD source and resolution for most types of filters, not an AR corrected "square pixel" resize - because you are scale the defect(s) larger thus making your job more difficult...

Clone tools in photoshop does better but 3000 frames of cloning is not to my liking.,

You can create a PS batch action to clone/heal/fill/inpaint - but all those variations on PS batch methods generally leave some slight flicker between frames because they are single image spatial repairs, not temporal video repairs. Often you need to follow up with some temporal smoothing over the area with a mask +/- regrain/renoise the area to match if it was blurred

would become by using previous good hor pixel(299) and hor pixel(306) on both side alternately to overwrite the bad pixels

The above looks simple enough to me. No blurring, no band lines, no random guessing based on fancy algorithms, etc.

How hard can it be?

If I understand your description, that should create vertical band artifacts if you alternate lines.

It sounds like you want to taking vertical column in x position 299 and alternating with vertical column in x position 306 to replace the defect area from the same frame ?

ie. Instead of vertical column in x pos = 299,300,301,302,303,304,305,306, where 300-305 are "bad", you get 299,306,299,306,299,306,299,306 . Those will generate displaced stripe artifacts

Previously when i mod, the original viewer (left vdub) stays the same and the processed viewer changes and therefore i can see the differences. For some reason your code changes both viewers and therefore both looked the same, despite the fact it worked, but gave me the impression that nothing was working.

When you load an avs script, that output of the avs script is the input into vdub, so left preview is the result of the avs script . The right is if you apply vdub filters afterwards

You can compare origina/filtered with an interleave script , where each frame switches back and fort when you advance .
interleave(original,filtered)

Or you can use avspmod to compare scripts in tabs. The number keys hot swap versions and there is a shared timeline. If you advance to frame 30, all versions will snap to 30, so it's easy to compare different versions of scripts

A simple question. Does avisynth have the code ability to copy a specific pixel and paste it in a specific place to replace another pixel? No blurring, no band lines, no random guessing based on fancy algorithms, etc.

x,y , or x column ?
A crude way is crop, stackhorizontal and/or stackvertical


Other options -

There are dedicated wire removal tools, or scratch removal in restoration programs, FX programs you might try

And there are other inpainting / logo removal tools similar to what John suggested that you can try. There are neural net/GAN inpainting methods available too

[coolgit]

function InCopy444(clip v, int X, int Y, int W, int H, float alpha)
{
    hzfill = StackHorizontal(v.Crop(X-1, Y, 1, H), v.Crop(X+W, Y, 1, H)).PointResize(W, H)
    vtfill = StackVertical(v.Crop(X, Y-1, W, 1), v.Crop(X, Y+H, W, 1)).PointResize(W, H)
    H > W ? Overlay(v, hzfill, X, Y, opacity=alpha) : Overlay(v, vtfill, X, Y, opacity=alpha)
}    

FFVideoSource("SH_2_secs.avi") 
ConvertToYV24() # must be YV24 or RGB to crop on odd pixel boundaries

InCopy444(last, 362, 2, 6, height-4, 1.0) # X, Y, W, H, strength (0=none, 1.0=full)
InCopy444(last, 577, 2, 6, height-4, 1.0)
InCopy444(last, 791, 2, 6, height-4, 1.0)

That didn't work very well.

Is there a complete manual on avisynth commands as when i searched for incopy444, i can't find it anywhere?

[coolgit]

*In general you want to use the original DVD source and resolution for most types of filters, not an AR corrected "square pixel" resize - because you are scale the defect(s) larger thus making your job more difficult...

Interesting. The first thing i did was deinterlaced and then resize to correct anamorphic 1024x576 using staxrip. I do without resize in the next episode.

Clone tools in photoshop does better but 3000 frames of cloning is not to my liking.,

You can create a PS batch action to clone/heal/fill/inpaint - but all those variations on PS batch methods generally leave some slight flicker between frames because they are single image spatial repairs, not temporal video repairs. Often you need to follow up with some temporal smoothing over the area with a mask +/- regrain/renoise the area to match if it was blurred

I used image processor (batch) to run actions (36 commands) to get where i am. Just left with thin white lines. Clone/heal in actions is a terrible idea. I used clone (click area i want and paste once, no brushing movement) to remove dust/spot/scratch on the first and last frames of scene changes and denoise those frames to the same level as the rest.

would become by using previous good hor pixel(299) and hor pixel(306) on both side alternately to overwrite the bad pixels

The above looks simple enough to me. No blurring, no band lines, no random guessing based on fancy algorithms, etc.

How hard can it be?

If I understand your description, that should create vertical band artifacts if you alternate lines.

It sounds like you want to taking vertical column in x position 299 and alternating with vertical column in x position 306 to replace the defect area from the same frame ?

ie. Instead of vertical column in x pos = 299,300,301,302,303,304,305,306, where 300-305 are "bad", you get 299,306,299,306,299,306,299,306 . Those will generate displaced stripe artifacts

No. See my earlier post. It would be 299,299,299,299,306,306,306,306

A simple question. Does avisynth have the code ability to copy a specific pixel and paste it in a specific place to replace another pixel? No blurring, no band lines, no random guessing based on fancy algorithms, etc.

x,y , or x column ?
A crude way is crop, stackhorizontal and/or stackvertical


Other options -

There are dedicated wire removal tools, or scratch removal in restoration programs, FX programs you might try

And there are other inpainting / logo removal tools similar to what John suggested that you can try. There are neural net/GAN inpainting methods available too

x column only like 299,299,299,299,306,306,306,306. This example is 6 pixel width, my white lines are 2/3 pixel. So 3 width max. Vertically an odd pixel may be out of place at close zoom 300% scrutiny, but on video unnoticeable.

I have tried some of them but when I watched some videos of how it done, the background is either green screen or nice clear same colour areas without any other objects in the way making removal easy. In my clip, not so.

During video editing i will use temporal smoothing to smooth out the faint bands and hope i get an acceptable result. So far its looking good.

[jagabo]

function InCopy444(clip v, int X, int Y, int W, int H, float alpha)
{
    hzfill = StackHorizontal(v.Crop(X-1, Y, 1, H), v.Crop(X+W, Y, 1, H)).PointResize(W, H)
    vtfill = StackVertical(v.Crop(X, Y-1, W, 1), v.Crop(X, Y+H, W, 1)).PointResize(W, H)
    H > W ? Overlay(v, hzfill, X, Y, opacity=alpha) : Overlay(v, vtfill, X, Y, opacity=alpha)
}    

FFVideoSource("SH_2_secs.avi") 
ConvertToYV24() # must be YV24 or RGB to crop on odd pixel boundaries

InCopy444(last, 362, 2, 6, height-4, 1.0) # X, Y, W, H, strength (0=none, 1.0=full)
InCopy444(last, 577, 2, 6, height-4, 1.0)
InCopy444(last, 791, 2, 6, height-4, 1.0)

That didn't work very well.

But it does exactly what you asked for. It infills by copying pixels from the left and right toward the center.

Is there a complete manual on avisynth commands

No. But you can start with:

http://avisynth.nl/index.php/Main_Page
http://avisynth.nl/index.php/Category:Internal_filters
http://avisynth.nl/index.php/External_filters

as when i searched for incopy444, i can't find it anywhere?

Because I just created it to show you that your requested method would not look good.

[coolgit]

function InCopy444(clip v, int X, int Y, int W, int H, float alpha)
{
    hzfill = StackHorizontal(v.Crop(X-1, Y, 1, H), v.Crop(X+W, Y, 1, H)).PointResize(W, H)
    vtfill = StackVertical(v.Crop(X, Y-1, W, 1), v.Crop(X, Y+H, W, 1)).PointResize(W, H)
    H > W ? Overlay(v, hzfill, X, Y, opacity=alpha) : Overlay(v, vtfill, X, Y, opacity=alpha)
}    

FFVideoSource("SH_2_secs.avi") 
ConvertToYV24() # must be YV24 or RGB to crop on odd pixel boundaries

InCopy444(last, 362, 2, 6, height-4, 1.0) # X, Y, W, H, strength (0=none, 1.0=full)
InCopy444(last, 577, 2, 6, height-4, 1.0)
InCopy444(last, 791, 2, 6, height-4, 1.0)

That didn't work very well.

But it does exactly what you asked for. It infills by copying pixels from the left and right toward the center.

It does but the colour isn't right. Target colour became lighter, not replace. Like copy red to replace green and get light green instead of red. Is it necessary to do stackvertical, it looks the reason colours are lighter.

Is there a complete manual on avisynth commands

No. But you can start with:

http://avisynth.nl/index.php/Main_Page
http://avisynth.nl/index.php/Category:Internal_filters
http://avisynth.nl/index.php/External_filters

as when i searched for incopy444, i can't find it anywhere?

Because I just created it to show you that your requested method would not look good.

Oh. I do have much much much to learn.

[jagabo]

function InCopy444(clip v, int X, int Y, int W, int H, float alpha)
{
    hzfill = StackHorizontal(v.Crop(X-1, Y, 1, H), v.Crop(X+W, Y, 1, H)).PointResize(W, H)
    vtfill = StackVertical(v.Crop(X, Y-1, W, 1), v.Crop(X, Y+H, W, 1)).PointResize(W, H)
    H > W ? Overlay(v, hzfill, X, Y, opacity=alpha) : Overlay(v, vtfill, X, Y, opacity=alpha)
}    

FFVideoSource("SH_2_secs.avi") 
ConvertToYV24() # must be YV24 or RGB to crop on odd pixel boundaries

InCopy444(last, 362, 2, 6, height-4, 1.0) # X, Y, W, H, strength (0=none, 1.0=full)
InCopy444(last, 577, 2, 6, height-4, 1.0)
InCopy444(last, 791, 2, 6, height-4, 1.0)

That didn't work very well.

But it does exactly what you asked for. It infills by copying pixels from the left and right toward the center.

It does but the colour isn't right.

The color is exaclty right. What you don't understand is that there is a lot of noise and discoloration in your source. When you copy an individual pixel to several adjacent columns those discolored pixels become much more visible. And edges become badly aliased. Here's a synthetic image that shows clearly what's happening:

StackHorizontal(BlankClip(width=32, height=64, color=$404040), BlankClip(width=32, height=64, color=$c0c0c0))
StackHorizontaL(last, last)
StackHorizontaL(last, last)
StackHorizontaL(last, last)
StackHorizontaL(last, last)
hzbar = last
StackVertical(last, hzbar.RGBAdjust(b=0.0))
StackVertical(last, hzbar.RGBAdjust(r=0.0))
StackVertical(last, hzbar.RGBAdjust(g=0.0))
StackVertical(last, hzbar.RGBAdjust(r=0.0, b=0.0))
StackVertical(last, hzbar.RGBAdjust(b=0.0, g=0.0))
StackVertical(last, hzbar.RGBAdjust(r=0.0, g=0.0))

ConvertToYV24()
Rotate(23.0)

InCopy444(last, 362, 2, 6, height-4, 1.0) # X, Y, W, H, strength (0=none, 256=full)
InCopy444(last, 577, 2, 6, height-4, 1.0)
InCopy444(last, 791, 2, 6, height-4, 1.0)

[Attachment 60088 - Click to enlarge]

[Selur]

@jagabo: Nice, catch with Dekafka.
I used:

ClearAutoloadDirs() # disable auto-loading
SetFilterMTMode("DEFAULT_MT_MODE", MT_MULTI_INSTANCE) # enable multi-threading
LoadPlugin("I:\Hybrid\32bit\AVISYN~1\LSMASHSource.dll") # http://avisynth.nl/index.php/LSMASHSource
LoadPlugin("I:\Hybrid\32bit\AVISYN~1\AGC.dll") # http://avisynth.nl/index.php/HDRAGC

# loading source: C:\Users\Selur\Desktop\SH_2_secs.avi
# color sampling YV12@8, matrix: bt709, scantyp: progressive, luminance scale: limited
clip = LWLibavVideoSource("C:\Users\Selur\Desktop\SH_2_S~1.AVI",cache=false,dr=true,format="YUV420P8", prefer_hw=0)
# current resolution: 1024x576

# color modifications to better see the blue line
clip = clip.HDRAGC()

org = clip # copy current state to 'org'

# filtering
clip = clip.ConvertToYV24(interlaced=false)
clip = clip.dekafka(363, 2, 8, clip.height-4, 256) # X, Y, W, H, strength (0=none, 256=full)
clip = clip.dekafka(578, 2, 8, clip.height-4, 256)
clip = clip.dekafka(792, 2, 8, clip.height-4, 256)
clip = clip.ConvertToYV12(interlaced=false)

# add label to clips
org = org.Subtitle("original")
clip = clip.Subtitle("dekafka")

Interleave(org, clip) # Interleave clips 

PreFetch(16) # set threads to use http://avisynth.nl/index.php/SetFilterMTMode
# setting output fps to 25.000fps
AssumeFPS(25,1)
#  output: color sampling YV12@8bit, matrix: bt709, scantyp: progressive, luminance scale: limited
return last

# custom functions
function dekafka(clip clip, int Xstart, int Ystart, int X, int Y, int Amount)
{
  ytop = Ystart
  ybot = Ystart + Y
  xleft = Xstart
  xright = Xstart + X
  topline = clip.Crop(Xstart, ytop-2, X, 2)
  bottomline = clip.Crop(Xstart, ybot, X, 2)
  leftline = clip.Crop(xleft-2, ytop, 2, Y)
  rightline = clip.Crop(xright, ytop, 2, Y)
  logosrc_hor = StackVertical(topline, bottomline).Blur(0, 1.58).BilinearResize(X, Y)
  logosrc_ver = StackHorizontal(leftline, rightline).Blur(1.58, 0).BilinearResize(X, Y)
  Amount2 = (Y>=2*X) ? 255 : 128*Y/X
  # Amount2 is small if X >> Y => logoscr_hor is dominant
  logosrc = Layer(logosrc_hor, logosrc_ver, "add", Amount2)
  clip = clip.Layer(logosrc, "add", Amount, Xstart, Ystart)
  return clip
}

to look at the results. (i used Y=8)
Dekafka really does a nice job on that clip.

Cu Selur

[coolgit]

The color is exaclty right. What you don't understand is that there is a lot of noise and discoloration in your source. When you copy an individual pixel to several adjacent columns those discolored pixels become much more visible. And edges become badly aliased. Here's a synthetic image that shows clearly what's happening:

I will try this later on my post photoshop-ed images as source when i merge images to video. The new video will have far less noise so maybe i get a better result.

[jagabo]

The color is exaclty right. What you don't understand is that there is a lot of noise and discoloration in your source. When you copy an individual pixel to several adjacent columns those discolored pixels become much more visible. And edges become badly aliased. Here's a synthetic image that shows clearly what's happening:

I will try this later on my post photoshop-ed images as source when i merge images to video. The new video will have far less noise so maybe i get a better result.

You may need to adjust the location and width of the infills. These values work better than the earlier script:

InCopy444(last, 362, 2, 6, height-4, 1.0) # X, Y, W, H, strength (0=none, 256=full)
InCopy444(last, 578, 2, 6, height-4, 1.0)
InCopy444(last, 791, 2, 6, height-4, 1.0)

[coolgit]

Tried the above on another file with similar problem and worked well.

The original source file we tried on had banding which made difficult. Without the banding it worked better.

However i'd discovered a roundabout way on another clip, camera fixed position, no motion and white to grey background with one ghastly line. Thankfully the line is on its own, no other objects. Found a way to use photoshop action by going into camera raw section, spot removal, mask line, change from heal to clone, to clone another area similar to the defect area, change back to heal - photoshop then removed the line and use heal from another area as guide. Not perfect but the whole scene didn't move thus no flickering, did 300 frames in one go and new heal section looked part of the background. Watching it on tv looks natural.