Scratchy white streaks on video encoded from old VHS

[Scott L]

Hi,

I have a lot of high bitrate DivX files which were sourced from old VHS tapes several years ago, but unfortunately, there is a recurrent defect in many of them in the form of a scratchy white streak that skips around horizontally (I've circled it on the frames shown below). In nearly all cases, the streak only occurs in the top right section of the picture. Throughout the duration of any particular video, the streak remains at exactly the same level, and only moves horizontally, generally not extending any further than to the middle of the picture. In VirtualDub, I've tried using the Logoaway filter to blur out the affected area, but regardless of the settings used, this results in a quite noticeable blur line, especially whenever there is vertical panning.

Does anyone here happen to know of a good program that would be capable of eliminating the white streaks specifically, without leaving any undesirable distortion (as is the case with the Logoaway filter)? If I am able to find a method of doing this, I plan to convert the DivX files to DVD format as they are restored.

Thanks in advance for any advice -

Scott

[guns1inger]

avisynth and descratch might be a place to start.

[Scott L]

Hi,

Thanks for your reply.

Despite having worked toward familiarizing myself with AviSynth and its DeScratch filter, I just can't seem to grasp working with a program that lacks a GUI, and have been unable to produce a suitable DeScratch script, even though the filter's documentation includes an example script.

If I provide a short video sequence (in DivX format) of the problem I'm trying to fix, would anyone here be willing to put together an appropriate DeScratch script for me? It would go a long way toward helping me to get started, and would be greatly appreciated.

Since the defect that I want to correct is quite specific and limited to only a small section of the picture, I probably only require a rather simple script that I can make minor adjustments to as needed for each video to be restored.

Thanks again -

Scott

[guns1inger]

Post it somewhere and I'm sure you will get several different solutions suggested.

[Scott L]

Ok, here's a 20 second sample -

Scott

[bendixG15]

I did look at your video, with the streaking UFO at 2:00 o'clock

Maybe if I update, then someone can offer some assistance.

[FulciLives]

Hi,

I have a lot of high bitrate DivX files which were sourced from old VHS tapes several years ago, but unfortunately, there is a recurrent defect in many of them in the form of a scratchy white streak that skips around horizontally (I've circled it on the frames shown below).

Would it be possible to go back and re-capture the VHS Videos? It looks like if you were to play the tapes back on a better quality VHS VCR that you might (and I do stress might) be able to eliminate the problem.

If anything you would end up with a better quality product working from a higher quality INTERLACED source. The big problem with MPEG-4 (DivX and XviD) is that you end up with DEINTERLACED video and that is BAD for image quality.

Anyways I'm downloading the supplied AVI file (sample.avi) and I'll play around with it and create a sample AviSynth script ... assuming I can get it to work 8)

- John "FulciLives" Coleman

[FulciLives]

Well I used the following AviSynth script:

LoadPlugin("C:\COPY_DVD\descratch0900\descratch.dll")
AviSource("L:\ZZZ\sample.avi")
ConvertToYV12()
TurnLeft()
DeScratch(mindif=4, maxgap=20, minlen=300, blurlen=50, keep=100, border=0, maxangle=0)
TurnRight()
LanczosResize(688,480,0,0,608,464)
AddBorders(16,0,16,0)

This "works" in that the script is a proper working script but it didn't do a damn thing as far as I could tell when it came to removing the artifact(s).

I'm thinking that DeSpot may be better. I'll give it a try next.

- John "FulciLives" Coleman

[FulciLives]

Well I tried DeSpot:

LoadPlugin("C:\COPY_DVD\Despot\avisynth_c.dll")
LoadCPlugin("C:\COPY_DVD\Despot\despot.dll")
AviSource("L:\ZZZ\sample.avi")
ConvertToYV12()
DeSpot(p1=18,p2=10,pwidth=128, pheight=10,sign=0,blur=4)
LanczosResize(688,480,0,0,608,464)
AddBorders(16,0,16,0)

This gave some decent results but didn't remove the artifacts 100% of the time.

I guess you just have to play around with the parm settings to get the right "combo" that works.

- John "FulciLives" Coleman

[Scott L]

After having tried the AviSynth scripts, I've concluded that doing re-captures from the VHS tapes is indeed the simplest and best solution to this problem.

Thanks again for the help -

Scott

[Bjs]

Do not forget , you may want to run those tape's through a cleaning device as used by video hire center's to remove dust / grit , that may have got inside over time while in storage ... this can cause problems for the playback device as noted .

[vhelp]

Evening Scott L,

I gave your inquiry a look'sees ..

I ran your "sample.avi" through a few tests. And, after analyizing it
for the issue you indicated, I saw that those "white streaks" seem to
be a consistant pattern. They do fluctuate from small -to- large
"white streaks", but this might be suitable for a specific AVIsynth
script that is user configured to [locate] those consisitant pattern
areas, and target only those areas with a removal (some kind of area-based
blend or convolution of the surounding pixels). If it were just one or
two frames, you could probably do that by hand. But it seems that almost
every frame consist of these streaks, and some are followed by and slight
"black" tail, either in front or behind the white streak.

The pattern I mentioned, looks sort of like a comment streaking across
the picture area in a Horizontal motion. It looks something like this:

.========----
.========----
.========----

or, from a matrix point of view:

[1][1][1][1][1][1][1][1][1][1][1]
[o][=][=][=][=][=][-][-][-][1][1]
[o][=][=][=][=][=][-][-][-][1][1]
[o][=][=][=][=][=][-][-][-][1][1]
[1][1][1][1][1][1][1][1][1][1][1]

And vary in length..

Where:

** "o" is the (black) head part, and
** "=" is the white streaks are consistantly white, and
** "-" is the gradient (white -to- gray -to- regular pixels) tail.
** "1" are the good surounding pixels, possibly to help replace/hide bad ones.
**
** And, there is THREE rows (or, 3x3 -to- 3x20 -to- 3x30 pixels blocks)
** give or take a few.

Now, if you *keep* your current AVI video of the prev project, and then
re-captured again, you -might- be able to use that (in an AVIsynth script)
to compare each frame and replace the effected areas with clean pixels
from -one of- the good frames, possibly without any area-base blend or
convolution step.

In any case, the script could be designed so that it only repairs according
to user defined parameters.., posibly like this:

repair(n)

Where:

** n is the value or line (row) that the white streaks begin.

It might take some more work to fine-tune the design of the script to
remove the bad pixels, either with a blend/convolution or a A/B picture
compare and replace routine. Some frames might have more than one white
streak, and/or may be on more than one row.

Of course, this might not work if those *bad* pixels are permanent in
every capture, and only a area-based or convolution (of just those areas)
will help repare (hide) them.

-vhelp 4208

[edDV]

Those "streaks" resulted either from tape dropouts (dust or holes in the tape emulsion) or from capture pixel drops. The "comet tail" looks more like a tape dropout. Hardware "dropout compensators" that work with interlace video substitute pixels from the previous line to mask the dropout. They maintain a line of memory from the previous field to provide pixels for substitution. A dropout is detected by a discontinuity in the RF read from tape.

When you deinterlace, you spread the defect to at least two lines, so the software filter must go to higher lines to find uncontaminated pixels to substitute. You can't just average since the discontinuity is large.

[vhelp]

Evening edDV,

The whole point of the *idea* was to try and hide or reduce the streaks
noticable distrations. If you average (whatever method) with surounding
cleaner pixels, then you help to reduce these streaky distractions.. even
if you don't have -as much- surounding pixels to work with, you have some-
thing to work with to reduce the distractions.

But, I don't know how important this (the work envolved) might be to this
user

But its a nice idea to consider. And I'm sure there are 1/2 a dozen more
out there.

Thank you for discribing the actual cause of this issue.

-vhelp 4210

[edDV]

Evening edDV,

The whole point of the *idea* was to try and hide or reduce the streaks
noticable distrations. If you average (whatever method) with surounding
cleaner pixels, then you help to reduce these streaky distractions.. even
if you don't have -as much- surounding pixels to work with, you have some-
thing to work with to reduce the distractions.

But, I don't know how important this (the work envolved) might be to this
user

But its a nice idea to consider. And I'm sure there are 1/2 a dozen more
out there.

Thank you for discribing the actual cause of this issue.

-vhelp 4210

I wasn't commenting on your solution. Tape dropouts have been a problem since the first video tape. Early line based TBC's included this dropout compensation feature to mask dropouts. It is similar to the Photoshop "Rubber Stamp" or "Clone" tool that was done a decade and a half later to much public acclaim.

I have yet to see a video filter that does this well. It should be a no brainer since patents have expired by now. The design challenge is to detect the target pixels for correction. I'd first target luminance discontinuity that differs from lines above and below. That would need to be restricted by a fireworks picture analysis .