g-force

Arnie.d,

here's my first crack at it.

ConvertToYV12()

########## set black level, adjust gamma, saturation/hue
Levels(15,1.2,240,0,255)
Tweak(sat=1.5,hue=-5)

########## prefilter (for obtaining motion vectors)
source = last
source.TemporalSoften(7,255,255,25,2).Repair(source,9)

########## get motion vectors
bw_vec1 = last.MVAnalyse(isb=true, delta=1,pel=2,sharp=1,overlap=4,idx=1)
fw_vec1 = last.MVAnalyse(isb=false,delta=1,pel=2,sharp=1,overlap=4,idx=1)

########## apply motion vectors
bw1 = source.MVCompensate(bw_vec1,idx=2,thSAD=400)
fw1 = source.MVCompensate(fw_vec1,idx=2,thSAD=400)

########## 1st degrain stage
inter = Interleave(bw1,source,fw1)
inter.Clense(reduceflicker=false).TemporalSoften(1,255,255,25,2)

########## limit 1st degrain stage
last.Repair(inter,9)

########## 2nd degrain stage
TemporalSoften(1,255,255,25,2)
SelectEvery(3,1)

########## temporal min/max sharpening stage
pmax = source.MT_Logic(bw1,"max").MT_Logic(fw1,"max")
pmin = source.MT_Logic(bw1,"min").MT_Logic(fw1,"min")
MT_lutxy(last,last.RemoveGrain(12,-1),"x y - 1.7 * x +",U=2,V=2)
MT_Clamp(pmax,pmin,2,2,U=2,V=2)

 

-G

Arnie.d said:

 

Anyway, I made a testclip. It's just a raw capture upscaled to anamorphic, no other filters were applied (also no averaging). It was captured using the default capture settings:

Movealong

And another one:

Leia

And the flyover, average of 3 captures:

Flyover

 

I welcome suggestions about the contrast, color, saturation, brightness etc. (or any other suggestions or ideas).

 

 Arnie,

these are looking nice so far. I'll play around with these and (hopefully) the 3 different sources you will post and get back to you on some suggestions.

BTW, it is so nice to see a source without so much telecine wobble! I also agree that the whites and blacks don't appear to be clipped, but I would still take Moth3r's suggestion of checking these objectively just in case.

Keep up the good work!

-G

Very minor update to the script today. Slightly better starfield retention, slightly better local jitter stabilization.

-G

Arnie.d said:

g-force said:

Arnie.d said:

I didn't test the median filter. I thought the temperalsoften was better?

 

 

It all depends on what type of noise you have. Only one way to tell really...

Any chance you could post the same clip from 3 different captures?

-G

 Does it have to be uncompressed or is mpeg2 ok?

mpeg2 should be just fine for what I have in mind!

thanks,

-G

Arnie.d said:

I didn't test the median filter. I thought the temperalsoften was better?

 

 

It all depends on what type of noise you have. Only one way to tell really...

Any chance you could post the same clip from 3 different captures?

-G

Arnie.d, I take it the median filter wasn't as good?

-G

Davis said:

Still have the lines. It’s not “DGIndex”- that is fine. It seems to be induced by AVIsynth itself. When I run the AVS file on the Star Wars video (not encoding at all, just playing it in Media Player), the horizontal interlace lines appear. Nothing I do in TMPEGenc takes them away….

Hey Davis, did you ever get this figured out? I'm at a loss, since there really is no way the script could be interlacing the video.

Is it possible that it is the DVD ripping software? Are you absolutely sure that you are using "force film" in DGindex?

The only other thing that I could think of is that some plugin is not the most recent version and is defaulting to interlaced video. Might want to check that you are using the most recent versions of the needed plugins.

-G

Max_Rebo said:

I hope that makes sense, if not I'll try to explain tomorrow.

I definitely understand what you are saying!

Max_Rebo said:

I've been thinking about it and I'm still not sure if a median filter would be better, in your example you made the assumption that only a third of pixels duffer from noise and that no one pixel suffers noise in two captures, which is a highly idealised situation.

I think the median method would be a good way avoiding bursts of static or dropouts that may not be reproduceable so only effect one capture but I can't see how it would help with the smaller random variations present in all analouge captures, when dealing with random variations taking the mean of the multiple values is surely the better way.

You are correct. But it will depend on what type of noise you have. The random variations you are used to with VHS are much smaller with LD. The biggest problem with LD is the noise introduced by such things as laser rot and dot crawl. These are not random variations that affect the entire frame equally, but rather isolated incidents that do not always occurr in the same place on every capture. I guess it all depends on the type of noise you have, and hopefully Arnie.d will try both!

Max_Rebo said:

will the scenechange parameter allow the middle capture to be excluded from the average? as this would not be useful for it's original purpose and the differences are measured relative to the middle capture, so if the middle capture had a significant problem then the other two would be excluded and the problem would remain.

You are correct. I didn't think of this situation. I could code up something that would replace the scenechange parameter to still average the 2 good frames.

Max_Rebo said:

what does it really mean by 'average pixel change'? is it a percentage? if 5-30 is appropriate for a change in scene isn't 25 too great a difference for supposedly identical frames I wouldn't expect 2 captures of the same frame to be different by more that 1-2%

The scenechange parameter is the maximum allowable average pixel difference in luma from one frame to the next. 25 may not be the best value especially if the captures are relatively noise free.

Orinoco_Womble said:

 

g-force said:

Also, there may be some benefit to throwing a mocomped-denoised source into the mix as well.

 

Sorry, didn't understand that bit at all. Umm, whats a mocomped-denoiser?

That was Moth3r's quote, not mine! A mocomped-denoiser is one that motion compensates objects in previous and next frames to their positions in the current frame and does some sort of filtering (median, average, fft etc.) on those frames.

-G

Arnie.d said:

But do you really think it will make a difference for the image quality and why? I'm not convinced.

 You just want to minimize ground loops so that no AC line noise is introduced.

-G

Arnie.d said:

g-force said:

Arnie.d said:

I'll give that one a try as soon as I get my new voltage convertor.

 

You know, I just thought of something. That voltage converter may be adding some noise to your captures due to the fact that the computer ground and the LD player ground are no longer the same. You may get better results if you can run your PC off of the voltage converter also. This may be possible if you flip the switch near the power input of your PC to 115 instead of 230. Just a thought.

-G

 I can't switch my PC to 115. But the LD player doesn't have ground (if I think I know what you mean by ground).

 Bummer. Yeah, it still has a ground, just not a chassis ground. Second best is to make sure the PC and the voltage converter are connected to the same outlet.

-G

Arnie.d said:

I'll give that one a try as soon as I get my new voltage convertor.

You know, I just thought of something. That voltage converter may be adding some noise to your captures due to the fact that the computer ground and the LD player ground are no longer the same. You may get better results if you can run your PC off of the voltage converter also. This may be possible if you flip the switch near the power input of your PC to 115 instead of 230. Just a thought.

-G

WXM said:

Darn, I was hoping you, Mother, or someone else knew a trick for easily wiping those dupe frames...

Mother and G-Force, would you say that your scripts above are better than the "Average" dll that I use? I would guess it would just be a different route to exactly the same result. What could the diffference be?

The doom9 forums are full of avisynth scripts to get rid of those dupe frames. Check it out! Luckily, the LD catures should give you exactly the same number of frames every time.

I would guess that the averaging script that Moth3r posted would give you EXACTLY the same results as whatever average you are using, as long as you give yours only 3 sources, so no worries! The median script though (yes Orinoco_Womble, it's the same concept as TOOT) should give even better results with 3 sources than an average of 3 sources, and a 3-source median may even be better than a 5-source average.

-G

resonator said:

As far as the upscaling to anamorphic is concerned you should definatly check out a small tool called "Video Enhancer" (www.thedeemon.com), it's the only resizing/upscaling method I know of that is temporal-based and gets detail and sharpness off the neighbouring frames. I've tried various upscaling algorithms (Lanczos, SSpline, RedGiant's InstantHD), and they all do pretty well but the "SuperResolution" does more than noticibly better. 30$ well spent for me. There's a version for AfterEffects available too, but it costs a 100.

 

I can think of a couple other temporal based sharpness filters, the best of which is currently used in my GOUT stabilization script. ;) I think it is important to temper people's expectations about "super-resolution". Super-resolution only works for things that are moving. This is a good thing for sources that suffer from DVNR smearing such as the GOUT, but for sources without smearing, all you're doing is getting rid of motion blur. Since all of the motion blur on the Pre-THX LDs is intentional, this would probably not be a good thing.

The best upsizers actually use Edge-Directed Interpolation. Arnie.d has already indicated that he plans on using NNEDI (Neural-Network Edge-Directed Interpolation) so I think he's already heading in the right direction.

-G