What are the best encoding profile options to reduce ghosting?

[jimbobuk]

Hiya..

I've asked about this over at shspvr but thought i'd come and ask here as well as it looks like i'm going to have to use my own profiles to tweak with this temporal filter..

I'm referring to the stuff in this page here

http://www.freytechnologies.com/2_papers/HPquality.txt

that i didn't even really know about.. I knew we could make profiles, i didn't realise we could add these temporal settings on top

disablefilter - 0:Disable, filter settings will be disabled.
medianfilter - 0:Disable, 1:Horizontal, 2:Vertical, 3:
Horz&Vert(default), 4:Diagonal
mediancoringlumahi - 0 to 255 (default 0)
mediancoringlumalo - 0 to 255 (default 255) mediancoringchromahi - 0 to 255 (default 0) mediancoringchromalo - 0 to 255 (default 255)
lumaspatialflt - 0:Disable, 1:1D Horiz, 2:1D Vert, 3:2D Horz & Vert Seperatable(default), 4:2D Non Seperable
chromaspatialflt - 0:Disable, 1:1D Horz(default)
dnrmode - 0:StaticBoth, 1:StaticTime/DynamicSpace, 2:DynamicTime/StaticSpace, 3:DynamicBoth(default) dnrspatialfltlevel - 0 to 15 (default 0) - used in static mode only dnrtemporalfltlevel - 0 to 15 (default 0) - used in static mode only dnrsmoothfactor - 0 to 255 (default 200) dnr_ntlf_max_y - 0 to 15 (default 15) - max NTLF Luma dnr_ntlf_max_uv - 0 to 15 (default 15) - max NTLF Chroma dnrtemporalmultfactor - 0 to 255 (default 48) - temporal filter multplier factor dnrtemporaladdfactor - 0 to 15 (default 4) - temporal filter add factor dnrspatialmultfactor - 0 to 255 (default 21) - spatial filter multiplier factor dnrspatialsubfactor - 0 to 15 (default 2) - spatial filter sub factor lumanltflevel - 0 to 15 lumanltfcoeffindex - 0 to 63 lumanltfcoeffvalue - 0 to 255 vimzoneheight - 0 to 15 (default 2)

Basically i know very little about this but i can only assume that my ghosting (most visible on characters with pans in dark scenes.. Star Trek episodes tend to really hilight it) is being caused by the temporal filter keeping too much of the previous frames behind.. it looks like you get a good 3 or so frames of edges back.. now do i switch to static temporal filter or do i tweak some of the other values to reduce this effect whilst keeping it in dynamic mode to best cope with everything.. I know roughly that the temporal filter helps the mpeg2 engine encode better so i'm nervous at say switching to static and a low filter amount in case it just messes everything up.. I'm also quite keen to add the tags to the profile so that dvd authoring software can better understand it.. I've been meaning to do this for ages but never took the plunge..

ANyways has anyone tweaked with this stuff.. I use DVD Standard at 3.5gig at the moment.. PAL.. i'll aim for a profile around this.. i'll probably modify the current standard one to suit my needs..

Thanks guys

[kny3twalker]

for dnr mode I like 3
but I never really messed it all the filters that much
it would be nice to see what other think
I would think it would nice to add some filters to custom profiles

[Opus4]

Are these other filters described someplace? I've seen them listed, but don't really know what they do.

- Andy

[jimbobuk]

someone needs to take the plunge and try a few of them i guess.. it'd be ideal to have a constant source so you could make captures of the same material and look at it...

It seems that no one really knows what they do exactly so changing them blindly could be a lot of fun but could also create some really crappy captures I imagine setting static temporal filter with a level of 0 would probably cause tiling in motion rather than ghosting... maybe there is no way to fix this but i will try sometime when i have the time...

Again if anyone has pioneered this and gone ahead and played with the values then please chime in with your results...

Am i the only person who notices this ghosting on most material... look for it in pans and again dark backgrounds... its really the only thing i can have a right to be slightly picky about my setup at the moment

[mls]

Oddly enough, there was a very long thread about the subject of "ghosting" that got started in an obscure section of the forum. You may find some of that info usefull:

http://forums.freytechnologies.com/f...&threadid=5906

Although there was a LOT of info given there, the end result for that persons problem was simply to adjust the color corection settings from within SageTV for each encoder card (instruction for that in another section of the forum that should already be linked to from the FAQ's).

If none of that info meets your needs, then altering the filter settings like you are thinking about may be the only alternative.

I will agree that there simply is not enough information currently available as to just what to expect from certain changes in those.

However, the really nice thing about SageTV is that you can add several different settings into the properties file and try them out. And, you don't have to make that the default for everything either. You could do a short manual recording with one setting, another with a different one, etc. and compare things until you find what settings best please you.

Of course, it would be nice to hear about what results people have with any of their tests here also...

[jimbobuk]

Thanks for the link, having read that I dont think its the same problem... I dont have permanent ghosting, only with motion.. its easier to see against a clear background... I've just scowered to try and find an example and have uploaded a video clip and various image sequences which i'll go through now to fully define the problem...

Before you say it, its not a problem with interlacing, its symptoms are similar in it requiring motion to occur but its visible on an interlaced tv via an MVP/burnt DVD and more importantly the latter set of images i've prepared proves its not.. more on that laters..

I've uploaded everything to here

http://jimtreats.com/pvr/

Ok first are some example shots of what i mean.. such as



You can make out 3 versions of his face.. though the 3rd furthest away is quite faint..

I have 3 other Battlestar Galactica sourced images along with some grabs from the clip that i've uploaded in this dir

http://jimtreats.com/pvr/pics/

They're a bit big and not that clear that its not interlacing so i'll move onto my better stuff..

Ok the clip sequence is 10 seconds long and still in mpeg2 frames straight from the card.. you can get that here

http://jimtreats.com/pvr/ghosting.zip

In the zip is the mpeg2 video, a dvd2avi made .d2v file along with the .d2a and .mpa audio ripped from it.. I've split it up like this as i've also included my .avs script that i used to analyse it further... thats in there as test.avs ... you'll need avisynth 2.5 installed and the required plugins but basically i've ripped the relevant images anyways so you dont have to (if you trust me that is, if not its there for you to recreate yourself)

Anyways the avs file looks like this

LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\MPEG2Dec3dg.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\Decomb510.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\mpasource.dll")
v = MPEG2Source("ghosting.d2v")
a = MPASource("ghosting MPA T01 DELAY 0ms.mpa", normalize=false)

AudioDub(v, a)
AssumeTFF().SeparateFields()
#Telecide(order=1, post=0, guide=2)
#FieldDeinterlace() #ConvertToRGB32()

There are the plugins for you if you want to get them.. anyways it simply loads in the video and audio seperately, dubs them together.. and then does a seperateFields() on the video.. Its this seperateFields thats key.. this quite simply pulls out the fields and puts them together.. so i go from PAL @ 25 full frames per second, to PAL @ 50 fields per second... This stuff is really neat you can look at field orders and see when the PVR has captured them unevenly.. anyways looking at this output in virtualdub (you can also just run the .avs file through windows media player or any other avi player) you can step through and no that interlacing is no longer in the video.. anything we see now has been captured wrongly... Here's a selection of images that are thankfully smaller, the full list is available here

http://jimtreats.com/pvr/pics/seperated_fields/

Can't remember this characters name but basically he's going to pull his head back and leave copies of his head ghosting behind him











And finally.. in the same clip is a more clearer one.. The camera is going to pan to the right leaving a smear of this guy, the final image i've enhanced (gamma-ed up) and you can really nicely see the profile of his face a few times











That's the problem.. and basically I'd love to fix it..

[sleight42]

Ah. I've observed this problem of late as well. A fix would be welcome, indeed. I believe that this problem is better characterized as "motion blur". Ghosting tends to refer to a consistant visual artifact that appears as a transparent shadow beneath objects on screen.

[mls]

You mention PAL... so that's an area I really don't know much about.

However, I do wonder if there may not be something similar to the 3:2 pulldown with NTSC that you might be seeing?

If you were to watch something recorded by Sage fed back out to a "standard" NTSC TV and step thru it frame by frame there would be 3 (reasonably) clear frames, then 2 blurred (sometimes with 2 different pics on top of each other slightly out of time from each), and then it all repeats again, 3 clean, 2 blurred, etc.

The 3:2 pulldown is needed to make 24 frames per second file come out right at 30 fps on NTSC TV's. I have no idea how they handle doing that sort of thing for PAL.

Anyway, just a thought. Don't know if it relates to what you're seeing or not.

[jimbobuk]

Cheers mls... I'm not sure what would cause your problems with 3:2 pulldown... you should only use this on material thats had the opposite encoding done on it (Telecine?) as you say its a way to get that framerate.. to be honest its something i dont like on US DVDs.. you can really notice it on horizontal pans...

Anyways in the UK, with PAL we're at 25fps... as such film material at 24fps is just played at 25fps... everything is slightly sped up, which in theory is an issue, but its all there.. simply progressive... its great for broadcast as well.. 90% of my recordings are all progressive anyways.. the problem is when the PVR starts recording mid frame which keeps it looking interlaced on a PC display but is fine on a normal tv.. (its also fine if you do some stuff with that avisynth style stuff)

Anyways.. this 3:2 isn't used at all on PAL... also i mentioned it above but the thin shots above are all done with fields seperated... there are NO interlaced fields present... also this motion blur as sleight wants to call it is a different effect than what you can get with interlacing...

Still no joy on testing these settings... i guess no one else has tried yet... when anyone does do.. report in on here

[kny3twalker]

is it definitely motion blur from watching a interlaced picture that has been deinterlaced
there are better decoders that can do this than others
Sonic and NVDVD do alot better of a good job than most
but the problem in myopinion is going to be that once you find a filtering that you are happy with for one type show
it will not do as good with others

This type of topic has been brought up many times
I would loveto see some custom recording qualities that are setup with certain filters for certain types of broadcast such as sports, scifi, drama

I think it would be a nice to see it happen
if anyone does mess with the filters for recording certain shows please post them

[jimbobuk]

Most of the images above are definately nothing to do with interlacing... the half height images are fields... ie. the components of a frame.. the things that are making up the odd and even lined based pictures... basically they should be progressive...

Only with some really subtle bob-ing do you get real interlacing issues.. they dont usually resemble this motion blur.. its usually much more pronounced... also remember these artefacts are being seen on dvds burnt out through an interlaced tv.. or via an MVP... there are no real interlacing issues when watched in this form..

what we are seeing is the motion blur from the temporarl filter keeping too much of the previous frame(s)...

There are issues with interlaced imagery as well.. to be honest now i have my MVP i doubt i'll be watching tv without going through it, unless i try to fire up my projector where it needs to be progressive... as i said before, being in the UK i am some what lucky that most broadcast stuff is inherently progressive.. of course tho most live stuff is all interlaced and subject to the problems you speak of... www.101fps.com or similar is all about that.. and there is stuff you can do to deinterlace well.. mostly tho its an offline task..

I aint going to have time to try this just yet, but when i do i'll be sure to report in any findings

[kny3twalker]

cool
thought we were referring simply to liveTV and recorded shows
not the whole burnt DVD and MVP thing

[jimbobuk]

I have seen the problem on tv.. i first spotted the battlestar galactica one whilst watching it on a normal tv.. beyond that though just be sure that those images above being individual frames are NOT problems from interlacing as showing them as frames totally removed interlacing from the equation.

[kny3twalker]

yeah I still think it is deinterlacing issue
http://www.100fps.com/
is the link and look at the second picture of the jetskier
I see this alot the examples you have from files created with the pvr 250

I also have seen pictures taken from an Xcard and you do not see these motion issues

[jimbobuk]

It's not deinterlacing

This'll be the 3rd time i've said this exact thing but basically when you seperate fields there is no longer any interlacing involved... interlaced frames are made up of two fields... the half height shots above and the video you get after running my .avs script included in the zip linked above is purely these fields.. they have been pulled out of the frames and seperated properly... read up on 100fps and you'll see that it talks about the kind of thing that seperating fields in avs is doing..

There is also this way: Displaying every field (so you don't lose any information), one after the other (= without interlacing) but with 50 fps. Thus each interlaced frame is split into 2 frames (= the 2 former fields) half the height.

This is what's happening.. its progressive...

As i've already said this though.. lets look at it another way..

Looking at your jetski example (i've put the image on my space so i dont leech his bandwidth)



You can see clearly 2 positions of the rider and the jetski.. looking closely you can obviously see each one being made by the alternate rows coming from the 2 fields.. importantly there is only 2 images.. this is all thats possible from interlacing as interlacing is only 2 frames... this is another reason to explain why this is not an interlacing problem.. looking at the brightened image from my pvr captures



You can clearly see 3 instances of the guys profile (1st the guy, 2nd the stronger outline of his face shape, 3rd the outer edge of his face) .. This is 3 frames being combined onto 1 field... even if you didn't believe me that a field is already different to a frame its still got one more image than is possible from interlacing alone..

Sadly understanding it don't really help you fix it.. its the temporal filter but its not clear if fixing it will not cause other issues.. such as the encoder just not being able to encode high quality without this blending from the temporal filter..

[flashbacck]

I see the same thing on my ehome wonder.

I'm surprised. The Pvr250 encoder's been out for ages, why hasn't this motion blur been fixed before? Is it a Sage thing?

[jimbobuk]

unlikely if not impossible to be a sage thing.. other than the presets having the wrong values set for each encoding profile.. I bet they're set to defaults recognised by Haugpauge (i'll never know how to spell that silly name

Sage is a frontend to the encoder really...

This motion blur to a point is required for mpeg2 to work i guess.. thats why its in there as a parameter for the encoding.. its this motion bluring that makes it easier to quickly encode frames following each other as being similar.. I dont properly understand it but to a point its required i think.. which is why i fear to fix this particular problem of ghosting I think we may introduce other things like bitrate saturation as the encoder can't cope ...

We've just gotta try things out and see i guess..

[mls]

Originally posted by flashbacck
I see the same thing on my ehome wonder.

I'm surprised. The Pvr250 encoder's been out for ages, why hasn't this motion blur been fixed before? Is it a Sage thing?

SageTV only tells the other software (encoder or decoder) what to record or what to play. Any video flaws being generated by the outside programs or hardware are not SageTV's fault. Nor, can SageTV correct for them (that again would be being done in outside software/hardware).

As for why this hasn't be fixed before... Probibly because most "average" people don't notice it, and those that do don't worry about it (or spend hours trying to understand all the technical details like some here are doing).

It's just NOT that a big of deal in my opinion. Sheeze, sit back, relax and enjoy TV... don't nit pit it to death pixel by pixel.

[jimbobuk]

you are right of course mls..

Its not THAT bad, but i did want to define it to hopefully find others who'd encountered it.. or more importantly others who had encountered it AND fixed it..

It is something you can fix with Sage, or at least try to by tweaking these temporal values..

I dont know whether its my tvs or not but it does seem particularly noticeable on certain material some of the time.. its just a niggle that i'd like to eliminate but agreed it aint a HUGE deal...

[kny3twalker]

I was not referring to that picture but rather this one


the statement at 100fps.com
says

But even after deinterlacing some red and some green pixels stay where the last field was.

after deinterlacing that is what you see
the first picture you posted was not deinterlaced but an interlaced image