I have a dvdr which was recorded from a vhs tape (which was a home TV recording). Video has a texture like a screen window mesh (only in motion, not if you pause the video). I dont think it's "dot crawl." Can anyone help me with some ideas how to improve it?
+ Reply to Thread
Results 1 to 23 of 23
By "mesh" do you mean the fine grain and noise? or do you mean the chroma noise ? or something else
I would save your money and use avisynth for free
# for shimmer removal # QTGMC(2,1,3,InputType=1,SLrad=2,sharpness=0.5)
If you want bonus points, you might notice some residual "shimmering" in the lamp in the top middle. You can use QTGMC's progressive mode (very slow to process) to treat that
Looks great, pdr. I think that -Vit- said that the latest QTGMC script will not accept number inputs like you have listed. I believe the latest script would need this:
if you play the clip on a TV, you can see the "mesh" I mean. I hesitate to call it grain since it's so prominent. Another analogy is half-tone newsprint (except again, the image isnt only made up of the dots as a newspaper image is).
Anyway, thanks Poison. The effect is minimized or corrected in both clips. The 2nd clip looks amazing. I'm hoping the first clip was perhaps a low rez version, and that I can do a better quality dvd encode with that script. I was never able to run QTGMC on my slow old PC. 'Course I might not be able to run the script, either, if my PC doesnt like it.
QTGMC can be made quite a bit faster and you can still gain most of the benefits that poisondeathray showed above. If you substitute a different preset in for his "Slower", you can realize considerable framerate increases without a lot of quality loss.
Here are fps numbers from my i5 M 540 Dell laptop running XP with 4GB ram:
"Slower" - 5.9 fps
"Faster" - 17.9 fps
"Ultra High" - 25.6 fps
I have enclosed a stacked clip with a few different settings. It looks like most of the grain is cleaned up just by QTGMC, so I eliminated both RemovedirtMC and LSFmod on these transcodes:
UL: Original processed only by inverse telecining [TFM().Tdecimate()]
UR: PDR's version minus RemoveDirtMC and LSFmod, including QTGMC
LL: Same as UR, but using only QTGMC("Faster",InputType=1) as the QTGMC string
LR: Same as UR, but using only QTGMC("Ultra High",InputType=1) as the QTGMC string
edit: Meant to ask pdr, according to the docs for QTGMC, SLrad should either be 0,1 or 3. What happens when it is set to 2?
But you're right, there is a double cleaning going on there (removedirtmc and QTGMC). I actually prefer removedirtmc for the occasional stains and dropouts that QTGMC misses (the side effect is removedirtmc denoises as well, just like QTGMC's side effect is as well) . I actually should have used noise retention settings for QTGMC if keeping removedirtmc . You can see the filesize is quite a bit smaller for #2 that I uploaded above - it's probably "overcleaned" and eroded too much detail. Anyways lots of options here and settings to play with
I was able to load QTGMC okay, but even at Preset=ultra high, dvd encode speed was only 7 frames per minute.
Using the script, the grain is removed but looks abit blocky (on TV), and some motion looked alittle unsmooth (maybe because of the blockiness). I guess you can see this in Poison's first avi.
I still think the QTGMC avi looks much better, even if there's some detail lost. I cant play mp4 on my dvd player, so I dont know how txporter's sample looks on TV.
I wonder if there's aspects of QTGMC found in other filters which would improve the video a tad without slowing down the PC?
Are you watching a MPEG2 or xvid AVI ?
By "blocky" do you mean the aliasing or macroblocking from encoding ? There is less aliasing in the 2nd one (look at the diagonal jaggies on the floor tiles)
The motion is the same in both . So I don't know what you're saying the motion is unsmooth ?
Often when videos are "too clean" this can cause macroblocking from high quantizers in dark areas like shadows and gradients. This is different than the aliasing issue. You can tone down the cleaning strength , add dither or add grain back
If you're doing a 2pass encode, use a lossless intermediate (so you endure the slow filters once instead of twice)
I'm not good at identifying and articulating video issues - could be both aliasing and macroblocking. I am encoding to dvd with HC (normal two pass) and watching the results on 30" TV. It's not as noticeable on my 12" monitor. I can see it around Werner's face when he's moving his head in the two-shot, drinking coffee. If I pause the dvd player, I can sometimes see the large blocks (macroblocks)? It's not too bad in that clip, but more noticeable in this other clip I was testing, so perhaps you can see what I mean. Used same script at best encode:
I see some blockiness, or straight edges, in the eyebrows of the shortest guy facing the camera. Also unsmooth motion as he moves his head slightly (not including the big motion turning his head at the end).
In the one-shot of Werner, you see blockiness in his upper lip. QTGMC didnt blur this stuff away (in the other clip) as I can still see detail in it.
Here's the untouched clip (not quite identical timecode) if you want to fiddle with it:
Last edited by spiritgumm; 1st Apr 2011 at 21:29.
It's actually both macroblocking (from low quality recording, low bitrate) and aliasing (matched fields already have the problem, so you don't get a clean IVTC) .
Normally you would IVTC this material, but you can use QTGMC in interlaced mode to bob deinterlace it , then srestore or some other decimation to get rid of the dupes . Deinterlacing will lower the overall quality more than if you IVTCed it, but in this particular case it might make sense to do so.
The "motion issue" is from frame displacement (x, y translation) ; you can use stab, depan or deshaker for that. You can use greyscale() to get rid of the green frame issue since you are using greyscale anyways to zap the chroma
MPEG2Source(operation demuxed.d2v", cpu=3)
#Lossless intermediate 2nd stage
mdata = DePanEstimate(range=1)
DePanStabilize(data=mdata, cutoff=0.5, damping=0.5, mirror=15, blur=30, method=1)
Greyscale #greenframe issue from depan
Thanks for all the work! However, with my slow PC & 512 Ram, will Qtgmc work any better in interlace mode?
And will the lossless line generate a huge file, or is it a temporary transition between the original and dvd encoding? It's a 90 minute movie.
I did try Stab before starting the thread (just with ivtc, greyscale, crop) but it didnt help then.
With such low ram , you probably want to divide it up into stages, because you'll run out of memory otherwise (avisynth needs to keep processed frames in the frame buffer, and it will balloon up with all those filters plus MPEG2 encoding) . So yes I would divide it up into 2 lossless stages, including 1 before the final DVD encoding (because you're probably doing a 2pass encode). They will be huge files, but you can use lossless compression (e.g. ut video codec, lagarith, huffyuv) .
I freed up some memory. Can someone outline the steps involved? I'm writing scripts in AvsP. Would I run part of the script, or all of it, in Vdub and compress to a lossless file? By two stages, do you mean using Trim to do the first half of the movie, and later the 2nd half? Or only processing part of the script (Qtgmc) in one stage, them the rest of the script in a 2nd stage?
How would I encode a lossless compressed file in HCenc?
I mean stages in the filtering of the script, because the it's the script processing that will consume memory, and stacking all those filters at once will not only consume too much memory, but also be much slower overall when you do 2 pass encodes in HCenc
where it says #lossless intermediate 2nd stage, everything above that goes into the 1st stage, everything below that gets applied to the 2nd stage in a 2nd script
In vdub, open the .avs script, use video=>fast recompress, video=>compression and select something like ut video codec (420), file=>save as avi
Then apply that 2nd part of the script to that exported AVI from vdub , by running that 2nd script through vdub, same steps
Then create a 3rd script to input into HCenc
Try to free up memory by closing non essential programs. Disconnect from the internet and close down firewalls, virus scans etc....
NeatVideo and many filters will ignore if it doesn't move. Try running NeatVideo at the same time as VDUb temporal smoother set to 2; chain the two filters in Vdub so the temp smoother runs first, and set NeatVideo's temporal filter to 2. If you see no effect at all, then the noise will likely have too little motion for most filters. which won't see the patterns as noise.
I have an old VHS that has the noise you describe. NeatVideo removed it almost completely, (you have to look hard to see any remnants). Depends on how thick the noise is.
Last edited by sanlyn; 20th Mar 2014 at 16:25.
That's an interesting suggestion about the source and solution for the grain. Looking at the reviews, I probably couldnt run NeatVideo on my current PC because of speed issues.
Speaking of which, I tried doing the script in stages (as described by Poison) but in stage 1, the speed was still extremely slow (looked like it would take 10 hours). I have a previous project I will probably have to do in stages, which I hope goes faster.
Meanwhile, the person I got the dvd from made another copy with a different dvd recorder. It wasn't corrupted, and the grain is less prominent (although he screwed up the beginning so I had to reencode part of the first dvd to splice into it).
So a relatively happy ending. Someday I'll get a new, faster PC...
Finding a better source is ALWAYS the better option !
I have NeatVideo on a home-built Athlon X64 2.4 GHz PC. NeatVideo runs at a typical 4fps on that machine, about 5fps on my old Gateway. NeatVideo is slow, period, but takes the place of several denoisers I no longer use. That amount of work takes time.
Sometimes I get noisy recordings on a machine, but a second try looks better. Only explanation I have is...some human error somewhere, or the phase of the moon, fixed by reconnecting and starting over ???
Last edited by sanlyn; 20th Mar 2014 at 16:25.
in this case, the video tape must have had the mesh/grain. The first dvdr was a full 4.37 gig dvdr from a Sony recorder, the 2nd was from a Panny in SP mode. The first was sharper and emphasized the grain, the latter less grainy.
Either way, Poison's QTGMC version still looks better. I dont know if it matters at this point, but I was only able to load part of it - QTGMC(preset="slow",border=true).