VHS (TV) with herzian echo

[lisztfr9]

Hi again,

here is a small example :

http://www.mediafire.com/?o4di938yjjl3kk5

I used Xvid in VD with high settings to record it.

A larger VOB file is also available here :

http://www.mediafire.com/?3ve5vytmnvk1181

However i won't use this part.

Thanks, L

[lisztfr9]

As it is, now :

http://www.youtube.com/watch?v=N-63KVL8AVQ

I lowered the fft3D filter :

fft3dfilter(sigma=3.0, plane=0, bt=3, bw=32, bh=32, ow=16, oh=16, ncpu=2, sharpen=1, interlaced=true)

Added Dehalo, turned to greyscale. Didn't crop.

It was too tedious to deal with the echo. I hope some guys will come out with the original print...

[lisztfr9]

Final version, until soon :

http://www.youtube.com/watch?v=1ZHZ7u_wkSc

- DeFlicker(scene=20, info=false)
- fft3dfilter(sigma=3.0, plane=0, bt=5, bw=32, bh=32, ow=16, oh=16, ncpu=2, sharpen=1, interlaced=true)
- FieldDeinterlace(blend=true)
- Cnr2()
- DeHalo_alpha().Greyscale
- Tweak(cont=1.1).Levels(0, 1, 255, 2, 235)
- Sharpen(0.16)

I realized that the exorcist from VD did i nice job, but it was due to it's ability to increase the contrast.. also removing chromanoise is necessary.

I choose to let fft3d only handle the luma, and cnr2 the chromnoise.

There is nothing much to improve on this bad source, every effect enhances the crap too. Remains the echo,

[lisztfr9]

I'v got an idea, but have not yet the skill to implement it :

Getting the mean luma near to the artifacts, and adjusting it over them. With Crop probably.

Thx, L

[pandy]

IMHO;

First remove HSync stripe - create (similar) overlay and subtract from video signal - this part is constant (static) and have very well defined structure.
Next step will be to remove ghost from video itself.

[lisztfr9]

I can't get it how to tile in Gimp (french Gimp, the translation is weird)... i would have try that, i'm searching. I read also about mask tools.

Got it. rtfm :)

[lisztfr9]

I managed to reduce luma on a line (6 column) near the center of the image, see left :

http://cjoint.com/12nv/BKbsHLh0asN.htm

But now i have 2 thin lines instead of that artifact being corrected !

The code therefore is :



# Crop slices

large = 6
start = 134
end = W - start

left = Crop(A, 0, 0, -end, 0)
right = Crop(A, start, 0, 0, 0)
Wr = width(right)

hband = Crop(right, 0, 0, -(Wr - large), 0)
right = Crop(right, large, 0, 0, 0)


B = BilinearResize(A, 6, H)

mt_lutf(B, hband, mode = "avg", expr = "y 10 -")

StackHorizontal(left, last, right)



IMHO i had to reduce the clip B (for INPOUT) to the same size as the stripe i wanted to repair, because of the mapping mechanism in mt_lutf...

So basically i wanted to take the average luma of the whole clip, and reduce the luma inside the stripe... but i have 2 small lines now... i keep running into problems !

L

Edit : maybe just reducing the size of the stripe to clean... anyway this cropping gave me a headache :)

Or cleaning the lines with Descratch...

Anyway this is far too simple, and not good on dark ground, so it should be more adaptive. The formula for y needs to be tweaked.

[pandy]

I think that HSync pulse is differentiated and for falling and rising edge there is kind of Gaussian pulse so IMHO your overlay HSync should be made as a gray field with level 50IRE, then you should crete white and black thin stripes in places corresponding to stripes on video and perform Gaussian blur on them to create you overlay mask, then blend mask with video.

For video itself i think you need perform similar operation, ie create mask that will be derivative (differentiate) video with some threshold... probably with help of tricky kernel (Sobel?), perhaps with some sharpening plugin (with motion detection?) that can show only map and map after Gaussian blur can be translated to anti-ghost mask...

(btw all i think need to be performed only on horizontal direction and i think that shift (delay) is very well defined due fact that HSync pulse have known parameters - width 4.7 +- 0.2(0.1 NTSC) uS - regularity (period) can be measured and offset calculated.)

Probably best method will be to use some specialized deconvolution plugin (is there anything for Avisynth to deconvolve self convolved video?)

[lisztfr9]

Hi Pandy,

You may be right with the overlay idea, at least i will get some blurring on the borders. With the mt_lutf i can't.

L