field differences ajustments

[smartel]

here is a source clip (.mpg) and a second one (.avi) with separatefields and smoothlevels for level adjustment

the uppêr field has a little better horizontal resolution ( i think)
and one is a little more purplish and the other a little more reddis

if you zoom in a little you can better see the letters on the keyboard , on the other they look scrambled

reduceflicker() may help and maybe mosquitoNR() too .. i want to know if a can cure this and not just throw away one field

[smartel]

my files are missing



[Attachment 74357 - Click to enlarge]
[Attachment 74358 - Click to enlarge]

[jagabo]

So you're looking to convert that MPG file into something more watchable? Something like this...

[smartel]

i was more thinking of a "edit friendly" version ( for Vegas movie studio)

the fields look to me like the colors are bleeding differently (top and bottom)

greenish on the top field and purplish all over

i guess that don't help deinterlacing

[jagabo]

Edit friendly is just a matter of how the final video is encoded and packaged. I used avc in MKV just to have a small file for upload. What did you think of the color stabilization and other cleanup? Do you want the file to be interlaced (like the source) or progressive (like my sample)?

[smartel]

i want both options , interlaced (for backup) and progressive (for upload and processing)

i have a few hours of those jams

if i can stay in YV12 all the way through , that would be a great thing

color stabilization is a good thing .. i usually monitor colors (hue) with colortools in VD ( i know it's in RGB)
but some of the JPSDR plugins work in YV12 and i do my script with Hybrid most of the time

and i don't know what other clean up you did ..bot looks like a good start

i want to try the mosquitoNR and for that i need a progressive source

if a deinterlace to double frame rate that would be easier to reinterlace after

[jagabo]

I did this to get the above video:

LWLibavVideoSource("VTS_01_1.mpg") 
ColorYUV(gain_y=-8, off_y=-12) # better levels, might be different for different clips
QTGMC(sharpness=0.5) # deinterlace
Santiag() # antialias MPEG DCT artifacts
MergeChroma(SMDegrain(thsad=50, tr=2, plane=0), SMDegrain(thsad=1000, tr=2, plane=3).aWarpSharp2(depth=20)) # mild temporal denoise luma, strong temporal denoise chroma
# encode 60p

An alternate if you want to retain the luma and only filter the chroma:

LWLibavVideoSource("VTS_01_1.mpg") 
ColorYUV(gain_y=-8, off_y=-12) # better levels, might be different for different clips
MergeChroma(Bob(0.0, 1.0), QTGMC().SMDegrain(thsad=1000, tr=2, PreFilter=4, plane=3)) # original luma, strong temporal denoise chroma
SeparateFields().SelectEvery(4,0,3).Weave() # re-interlace
# encode 30i

This leaves all the DCT artifacts in the luma.

[smartel]

MPEG2Source("H:\FNM\VIDEOHELP\VTS_01_1_temp\VTS_01 _1.d2v",idct=5)
ReduceFlicker(strength=2) #helps with wondering color fog
SmoothLevels(gamma=0.83,input_high=255,TVrange=fal se,Ecurve=0,HQ=true)
QTGMC(preset="Fast")
santiag(strh=2, strv=2, type="sangnom", nns=1, nsize=6, aa=48, aac=24, scaler_post="Spline36Resize").Turnleft()
Santiag(strh=2, strv=2, type="sangnom", nns=1, nsize=6, aa=48, aac=24, scaler_post="Spline36Resize").TurnRight()
MergeChroma(SMDegrain(thsad=50, tr=2, plane=0), SMDegrain(thsad=1000, tr=2, plane=3).aWarpSharp2(depth=20))

With Smoothlevels i'm dithering early on ..maybe that could go at the end
i'm not really familiar with MergeChroma and SMDegrain

[jagabo]

SMDegrain() is a motion compensated temporal noise reduction filter.

http://avisynth.nl/index.php/SMDegrain

Temporal meaning that it blurs away noise across successive frames, but not spacially within frames. Motion compensated means that it recognizes portions of the picture that have moved and blurs relative to that movement, not just the static location of pixels in each frame.

thsad is the strength of the noise reduction. Higher values remove more noise but can lead to blurring away details and/or ghosting of moving objects. tr is the temporal radius -- the number of frames before and after the current frame over which the noise is averaged. So tr=1 will average the pixel of the current frame, one frame before, and one frame after. tr=2 will average the pixel from the current frame, two frames before, one frame before, one frame after, and two frames after. Etc. plane=0 means only the luma plane is de-noised. plane=3 means the two chroma planes are de-noised.

MergeChroma() simply joins together the luma (greyscale image) of one video with the chroma (colors) of another video.

http://www.avisynth.nl/index.php/Merge

In my script MergeChroma() is used to combine a lightly denoised version of the video's luma with a highly denoised version of the video's chroma.

The line:

MergeChroma(SMDegrain(thsad=50, tr=2, plane=0), SMDegrain(thsad=1000, tr=2, plane=3).aWarpSharp2(depth=20)) # mild temporal denoise luma, strong temporal denoise chroma

could also be performed by two successive calls to SMDegrain:

SMDegrain(thsad=50, tr=2, plane=0) # mild temporal denoise only the luma (chroma unchanged)
SMDegrain(thsad=1000, tr=2, plane=3).aWarpSharp2(depth=20) #strong temporal denoise, and strong sharpening, of only the chroma (luma unchanged)

[smartel]

Getting back to my first question ..

If i wanted to filter at field level .. selecteven(). filter ; selectodd().filter...

would temporal filtering at 1 frame (even or odd field) before and 1 after have any benefit ? are the drawbacks that bad ?

this is a bad source and just hoping that this filtering could help the deinterlacing

i used reduceflicker because is field-aware .. can i make a filter that isn't .. be ?

like preparing the source for deinterlacing ..

sounds like i'm asking for a reduceflickermod

[lollo]

If i wanted to filter at field level .. selecteven(). filter ; selectodd().filter...
would temporal filtering at 1 frame (even or odd field) before and 1 after have any benefit ? are the drawbacks that bad ?

Applying a temporal filter after a SelectEven() / SelectOdd() for an interlaced video is wrong.
The original architecture is:

interlaced frames
frame1	        frame2  	frame3	        frame4
A	        B	        C	        D		(field 0) even lines
a	        b	        c	        d		(field 1) odd lines

each field is unique about content in time.

if you apply SelectEven().filter and SelectOdd().filter you are applying a filtering across an inconsistent time progression

field1Even	field2Even	field3Even	field4Even
A	        B	        C	        D		(field 0) even lines

field1Odd	field2Odd	field3Odd	field40dd
a	        b	        c	        d		(field 1) odd lines

because you are using a temporal filtering across consecutive fields that are not consistent (A-B-C-D or a-b-c-d).

To properly processing an interlaced video with temporal filtering you need to lossy deinterlace the original source and (eventually) interlace back at the end

[smartel]

So mosquito noise is a DCT artifacts.. and i have to use an antialiasing filter to smooth them out or a deblocker ..
and find a balance not to smooth out details .. would resizing help for this ..after the deinterlacing ?