x264/Handbrake deblock values -5 - +6, what do they mean?

[premiumcapture]

The deblock numbers go negative to -5 and positive to +6. What is the difference and what is gained by going negative?

[poisondeathray]

alpha and beta inloop deblock values. In short, negative values are weaker, positive values are stronger produce more blurry results

i.e Negative values lose less detail, if adequate bitrate is available . In low bitrate ranges, relative to content complexity, it will produce worse results

Read this thread for more info
http://forum.doom9.org/showthread.php?t=109747

[hello_hello]

Negative numbers can give you more picture detail but also increase the chance of "blocking". Positive numbers can blur a little more but reduce the chance of blocking. http://en.wikipedia.org/wiki/Deblocking_filter

A very exaggerated example of what blocking might look like. http://sp.cs.tut.fi/mobile3dtv/impaired-videos/blockedge30-r.png

The x264 tuning "film" sets deblocking to -1 and -1, although Handbrake doesn't show you the change in it's GUI. The way I understand it, back in the early x264 days (before the psychovisual enhancements were added), it wasn't as good at retaining detail, so people used the lowest deblocking setting they could get away with, without noticeable blocking. -3 and -3 seemed pretty common. These days, it's probably not necessary, and sticking to Tune Film is fine for most video. Tune Film also sets --psy-rd 1.00:0.15 (the default is --psy-rd 1.00:0.00). I don't think it changes any other settings.

For the record:
Tune Film = deblock -1, -1 and --psy-rd 1.0:0.15
Tune Grain = deblock -2, -2 and --psy-rd 1.0:0.25 and --aq-strength 0.5 and --ipratio 1.1 --pbratio 1.1 and --qcomp 0.8
Tune Animation = deblock 1, 1 and --psy-rd 0.40:0.0 and --aq-strength 0.6 and increases the number of B and Ref frames

I'm pretty sure the above is correct, but don't hold me to it. They're the changes from the default settings when using the corresponding tuning.

[premiumcapture]

Are these numbers related to banding in any way?

[LigH.de]

Banding is mainly related to too coarse quantization due to too low bitrate. Adaptive Quantization is a more common way to fight banding by spending more bitrate to low detail areas.

Positive numbers for deblocking parameters may be able to filter banding a bit more, but will also blur the whole picture where quantization is coarse. They should only be raised when there is not much structure and detail to be expected, specifically in classic cartoons/anime. Negative values may be useful to retain exaggerated film grain (like in "300") but will require a high bitrate, preferably a CRF of at most 15.

Please note that deblocking parameters are a more or less symmetrical feature. They change the behaviour of both encoder and decoder.

[Selur]

in general: no
banding is more a problem of rounding errors due to calculation restriction, that's why 10bit encoding helps a lot with banding (2 more bits to represent numbers)
-> either use a higher calculation bit depth or apply dithering to your source

[premiumcapture]

So select adaptive quantinization + lower rf value = less banding.

It sounds like most non-animated sources should be -1, -1.

If I choose to dither, should it only be in extreme cases or is it something I can always leave on?

I am primarily using Handbrake, TMPG Mastering Works 5, and I have played with MeGui. I can use the command line if necessary but I prefer to avoid when possible.

[hello_hello]

Whenever I have banding problems I lower the CRF value (I generally encode using CRF18 and lower it to CRF17 or CRF16 to reduce banding). Usually I wouldn't encode the whole video at a lower CRF value, just the section(s) where it's required. I do that by splitting an encode into sections and append the encodes together later. Or you can probably specify zones for the x264 encoder but I generally just split the encoding job. It requires adding --stitchable to the command line to ensure you can append the encoded video (and using the same x264 settings each time, although you can change the CRF value). I'm not sure how you'd do that using Handbrake. It's not too hard with MeGUI (it has an AVS Cutter under the Tools menu).

My logic is this.... if changing an x264 setting encodes a video more accurately, it invariably increases the bitrate too. Lowering the CRF value does the same, so rather than messing around with x264 settings all the time.....

Noise removal can increase the chance of banding because noise helps prevent it. Adding noise (dithering) after noise removal can help prevent/reduce banding also. There's Avisynth plugins for reducing any existing banding in the source video and plugins to help prevent the re-encoding process causing it. http://forum.doom9.org/showthread.php?p=1386559

MeGUI makes it fairly easy to modify the scripts it creates (add extra filtering manually etc). Using Avisynth plugins isn't hard once you understand the basics of creating scripts, which MeGUI can help you learn by looking at the scripts it creates and/or modifying them. MeGUI only applies "basic" Avisynth filtering for you. Anything else you need to download, install and add to a script manually yourself.

[hello_hello]

So select adaptive quantinization + lower rf value = less banding.

Adaptive quantisation is enabled by default. There's three settings. --aq-mode 0 = off, --aq-mode 1 = normal, and --aq-mode 2 = auto.

http://mewiki.project357.com/wiki/X264_Settings#aq-mode
I don't know if mode 2 is better for reducing banding. Or if it can have adverse effects. I've not really experimented with it myself.

http://mewiki.project357.com/wiki/X264_Settings#aq-strength
The tune grain and tune animation presets reduce the aq strength a bit (--aq-strength 0.5 and --aq-strength 0.6). The film tuning uses the default strength (--aq-strength 1.0).

Outside of those settings I haven't played with AQ Strength either. No doubt others have......

[premiumcapture]

Whenever I have banding problems I lower the CRF value (I generally encode using CRF18 and lower it to CRF17 or CRF16 to reduce banding). Usually I wouldn't encode the whole video at a lower CRF value, just the section(s) where it's required. I do that by splitting an encode into sections and append the encodes together later. Or you can probably specify zones for the x264 encoder but I generally just split the encoding job. It requires adding --stitchable to the command line to ensure you can append the encoded video (and using the same x264 settings each time, although you can change the CRF value). I'm not sure how you'd do that using Handbrake. It's not too hard with MeGUI (it has an AVS Cutter under the Tools menu).

My logic is this.... if changing an x264 setting encodes a video more accurately, it invariably increases the bitrate too. Lowering the CRF value does the same, so rather than messing around with x264 settings all the time.....

Noise removal can increase the chance of banding because noise helps prevent it. Adding noise (dithering) after noise removal can help prevent/reduce banding also. There's Avisynth plugins for reducing any existing banding in the source video and plugins to help prevent the re-encoding process causing it. http://forum.doom9.org/showthread.php?p=1386559

MeGUI makes it fairly easy to modify the scripts it creates (add extra filtering manually etc). Using Avisynth plugins isn't hard once you understand the basics of creating scripts, which MeGUI can help you learn by looking at the scripts it creates and/or modifying them. MeGUI only applies "basic" Avisynth filtering for you. Anything else you need to download, install and add to a script manually yourself.

I am currently working on batch encoding my video collection, so while if I was working to master a single title, this would certainly be the best approach, however, it sounds like the best thing to do would be to use higher RF values. I didn't know if the psychovisuals had anything to do with it and if turning everything on/off would be better, but MPEG-2 doesn't seem to suffer the same issue quite as often.

I recently encoded Avatar and the file went from 40gb -> 8gb. Certainly not as detailed but still looked excellent. Perhaps I have been doing this too long, but in dark scenes with little detail I see blocking not present on the original and it bothers.

[jagabo]

in dark scenes with little detail I see blocking not present on the original and it bothers.

8 bits of Y, U, and V (especially with the limited video range) are not sufficient to produce smooth gradients in the absence of noise. This is why you want to leave noise in the video by using a higher aq-strength (1.8 or so), or lower CRF. Some examples using --tune:

https://forum.videohelp.com/threads/359041-Is-the-film-preset-in-x264-ripbot-only-for-2...=1#post2270065

[premiumcapture]

in dark scenes with little detail I see blocking not present on the original and it bothers.

8 bits of Y, U, and V, are not sufficient to produce smooth gradients in the absence of noise. This is why you want to leave noise in the video by using a higher aq-strength (1.8 or so), or lower CRF.

Is this a result of the conversion process or just the nature of H.264? I certainly see the value of 10-bit video, but if the original is at 8-bit YUV and does not have the issues, is it due only to higher bit rate?

[jagabo]

Is this a result of the conversion process or just the nature of H.264?

It's the result of only using 8 bits. Even 8 bit (per channel) RGB is insufficient to produce smooth gradients in the absence of noise. Make some test videos with shallow, perfectly smooth, 8 bit YUV gradients then use tool like CSamp to read RGB values off the screen while displaying the video. You'll see that RGB values often jump by 2, 3, or even more between bands.

[hello_hello]

premiumcapture,
I'll confess I have problems with banding at times but banding/blocking just in dark areas, not so much. I wonder if that's due to TV/Monitor calibration?

Out of curiosity, as I happen to be encoding these at the moment, how much blocking do you see in the little sample below? It was noise filtered, but as it was fairly noisy to begin with I didn't apply dithering (usually I would).

Under normal viewing conditions, I can't see any obvious blocking/banding, except for where the light is shining on the book, but that looks about the same as the source anyway. The rest of the video.... there's blocks for days in dark areas, and large sections of single shades, but under normal viewing conditions I can't see them. If I crank the gamma up a bit or view the video on the PC monitor without expanding the levels to full range it all becomes easy to see and doesn't look all that good, but under normal viewing circumstances it looks fine. Obviously the picture quality isn't great, but that aside....

[Mephesto]

The deblock numbers go negative to -5 and positive to +6. What is the difference and what is gained by going negative?

First setting is deblocking amount to apply and second is deblocking threshold. The difference is really minimal though and there's little reason to use non-default values. For cartoons it makes sense to +1 +1 because it gets rid of some jaggedness on the edges caused by blocking but increasing any more may negatively affect sharp, legitimate details in the animation which is why it's largely avoided.

For films many use -1 -1 for an artificially sharper image.

What you should know is that the deblocking algorithm is automatically tuned to the level of quantization. If you use a super high bitrate, even a +6 +6 setting won't visibly smooth the picture because there's no blocks to fix and -6 -6 won't make anything look "worse" as there is no visible blocking to begin with.
Don't get too obsessive over the setting. Rule of thumb is -1 -1 for films and +1 +1 for cartoons. 0 0 if you aren't sure.

[premiumcapture]

premiumcapture,
I'll confess I have problems with banding at times but banding/blocking just in dark areas, not so much. I wonder if that's due to TV/Monitor calibration?

Out of curiosity, as I happen to be encoding these at the moment, how much blocking do you see in the little sample below? It was noise filtered, but as it was fairly noisy to begin with I didn't apply dithering (usually I would).

Under normal viewing conditions, I can't see any obvious blocking/banding, except for where the light is shining on the book, but that looks about the same as the source anyway. The rest of the video.... there's blocks for days in dark areas, and large sections of single shades, but under normal viewing conditions I can't see them. If I crank the gamma up a bit or view the video on the PC monitor without expanding the levels to full range it all becomes easy to see and doesn't look all that good, but under normal viewing circumstances it looks fine. Obviously the picture quality isn't great, but that aside....

I can see blocking around the person wearing the cross. It looks similar to noise, but perhaps its due to x264 but it looks more blocky to me than just noise. Clearly if one isn't looking for them, they probably won't see them, but I've been working with VHS for so long that I seem to pick up more than the average bear. Certainly a very watchable sample. I didn't notice banding.

I am using an ASUS 27" 1600x2560 IPS monitor with a Macbook Pro Retina and Windows 8.1, never calibrated but also never felt the need.

[premiumcapture]

The deblock numbers go negative to -5 and positive to +6. What is the difference and what is gained by going negative?

First setting is deblocking amount to apply and second is deblocking threshold. The difference is really minimal though and there's little reason to use non-default values. For cartoons it makes sense to +1 +1 because it gets rid of some jaggedness on the edges caused by blocking but increasing any more may negatively affect sharp, legitimate details in the animation which is why it's largely avoided.

For films many use -1 -1 for an artificially sharper image.

What you should know is that the deblocking algorithm is automatically tuned to the level of quantization. If you use a super high bitrate, even a +6 +6 setting won't visibly smooth the picture because there's no blocks to fix and -6 -6 won't make anything look "worse" as there is no visible blocking to begin with.
Don't get too obsessive over the setting. Rule of thumb is -1 -1 for films and +1 +1 for cartoons. 0 0 if you aren't sure.

So depending on the bit rate, this may not be necessary? Also, the -1, -1 is similar to unsharp mask in a way?

[poisondeathray]

It does NOT sharpen. It reduces the amount of deblocking / blurring . Reducing the amount the inloop deblocking is NOT the same thing as adding sharpen filters

The sharper, more noise, more dithering you have, the more bitrate you require to maintain a certain level of quality

It's not necessary to adjust the deblock filter at all when using typical bitrate ranges, in most scenarios

[premiumcapture]

It does NOT sharpen. It reduces the amount of deblocking / blurring . Reducing the amount the inloop deblocking is NOT the same thing as adding sharpen filters

The sharper, more noise, more dithering you have, the more bitrate you require to maintain a certain level of quality

It's not necessary to adjust the deblock filter at all when using typical bitrate ranges, in most scenarios

So if going from DVD, where blocks aren't necessarily noticeable though are sometimes, will the blocking get worse than the DVD if the negative values are too high (-6?)

[poisondeathray]

It does NOT sharpen. It reduces the amount of deblocking / blurring . Reducing the amount the inloop deblocking is NOT the same thing as adding sharpen filters

The sharper, more noise, more dithering you have, the more bitrate you require to maintain a certain level of quality

It's not necessary to adjust the deblock filter at all when using typical bitrate ranges, in most scenarios

So if going from DVD, where blocks aren't necessarily noticeable though are sometimes, will the blocking get worse than the DVD if the negative values are too high (-6?)

Nearly all DVD's have fine macroblocking to begin with if you look closely.

At a given bitrate, yes blocking will be worse with more negative values

You're probably not using average bitrate based encoding (you're probably using CRF encoding), so this has less relevance. You cannot compare encodes at different bitrates. So is you encode at CRF say 16, and compare the effect of changing deblocking values, it's not relevant because the final encoded bitrate will have changed .

[premiumcapture]

It does NOT sharpen. It reduces the amount of deblocking / blurring . Reducing the amount the inloop deblocking is NOT the same thing as adding sharpen filters

The sharper, more noise, more dithering you have, the more bitrate you require to maintain a certain level of quality

It's not necessary to adjust the deblock filter at all when using typical bitrate ranges, in most scenarios

So if going from DVD, where blocks aren't necessarily noticeable though are sometimes, will the blocking get worse than the DVD if the negative values are too high (-6?)

Nearly all DVD's have fine macroblocking to begin with if you look closely.

At a given bitrate, yes blocking will be worse with more negative values

You're probably not using average bitrate based encoding (you're probably using CRF encoding), so this has less relevance. You cannot compare encodes at different bitrates. So is you encode at CRF say 16, and compare the effect of changing deblocking values, it's not relevant because the final encoded bitrate will have changed .

I've found that the luma is done pretty well at decent releases and don't see much blocking, but chroma can get a bit blocky. If DVDs are inherently blocky, then softening isn't such a bad thing then. It sounds like one cannot have block free video and 1:1 sharpness.

[poisondeathray]

I've found that the luma is done pretty well at decent releases and don't see much blocking, but chroma can get a bit blocky. If DVDs are inherently blocky, then softening isn't such a bad thing then. It sounds like one cannot have block free video and 1:1 sharpness.

Look closely, and you will see macroblocking on virtually any retail DVD release. Unless there is no motion. It's the nature of MPEG2. Even at 2-3 the legal DVD bitrates you would see macroblocking

Typically you don't adjust encoder settings to reduce source macroblocking. That's a very poor way to do it. You filtering before encoding

If source is clean (really clean, like studio master), then you dither and possibly add grain before encoding. This is what studios do before releasing

[premiumcapture]

I've found that the luma is done pretty well at decent releases and don't see much blocking, but chroma can get a bit blocky. If DVDs are inherently blocky, then softening isn't such a bad thing then. It sounds like one cannot have block free video and 1:1 sharpness.

Look closely, and you will see macroblocking on virtually any retail DVD release. Unless there is no motion. It's the nature of MPEG2. Even at 2-3 the legal DVD bitrates you would see macroblocking

Typically you don't adjust encoder settings to reduce source macroblocking. That's a very poor way to do it. You filtering before encoding

If source is clean (really clean, like studio master), then you dither and possibly add grain before encoding. This is what studios do before releasing

You are saying noise and grain are not the enemy? Heresy! Just kidding...

I am looking to convert DVDs and Blu-Rays to smaller sizes while maintaining close quality, but the more I notice the output, even at 0.5x-0.7x filesize, the difference to me is huge. I will not go through the trouble of all this filtering for commercial releases, just because its easier to keep it in commercial format, but it shocks me sometimes how much is lost in the process, not that the resulting video is absolutely terrible, but the fine details never seem to be absolutely replicated. Perhaps I am a perfectionist, but looking at my 100+ disc collection, I do not want to spend more time converting than I would actually spend watching the video.

I have yet to find a perfectly cure all solution for this, though I will say that early HEVC builds show some promise for my old SD material.

[poisondeathray]

If you're a really a perfectionist, you wouldn't convert them. 1:1 backup. 4TB HDD are ~$150 now. Don't bother wasting time, electricity, killing trees

(...On the other hand, many retail DVD's are chalk full of flaws, and if you really were a "perfectionist", you would fix them. )

[premiumcapture]

If you're a really a perfectionist, you wouldn't convert them. 1:1 backup. 4TB HDD are ~$150 now. Don't bother wasting time, electricity, killing trees

(...On the other hand, many retail DVD's are chalk full of flaws, and if you really were a "perfectionist", you would fix them. )

Not so much about desktop viewing, using a combo of MakeMKV for NTSC releases and DVD Decrypter for difficult PAL releases, I have everything backed up 1:1, more about taking them on my tablet, which is a Samsong Tab S, with a QHD screen. I have a 64gb card and usually just convert with intel quicksync @ 720p for blu rays and default quick sync for DVDs, but I wanted to make a portable secondary library.

I agree most DVDs have flaws as stated, though I am of the belief that in many cases, if the flaws are part of the original presentation, i prefer to keep them. VHS on the other hand, being flawed by nature, is the bane of my video restoring existance

[poisondeathray]

I agree most DVDs have flaws as stated, though I am of the belief that in many cases, if the flaws are part of the original presentation, i prefer to keep them. VHS on the other hand, being flawed by nature, is the bane of my video restoring existance

I would argue those flaws are not as the director or producer intended . They are limitations from being produced on DVD. I'm primarly talking about macroblocking from MPEG2. DVD , like VHS is also flawed by nature, albeit to a lesser degree. (and yes there can be lots of other DVD issues like cadence problems, field blending etc.... )

eg. Any action scene in a DVD will show macroblocks. You want to keep those flaws in there ? Do you think the director wanted to look like that? Just go to the same BD source (if available), and do you think those flaws exists (to the same degree)? The problem is MPEG2 and DVD bitrates. Sure, in motion, those issues are less visible, but I would argue any perfectionist wouldn't want that. I guess it's your choice

[premiumcapture]

I agree most DVDs have flaws as stated, though I am of the belief that in many cases, if the flaws are part of the original presentation, i prefer to keep them. VHS on the other hand, being flawed by nature, is the bane of my video restoring existance

I would argue those flaws are not as the director or producer intended . They are limitations from being produced on DVD. I'm primarly talking about macroblocking from MPEG2. DVD , like VHS is also flawed by nature, albeit to a lesser degree. (and yes there can be lots of other DVD issues like cadence problems, field blending etc.... )

eg. Any action scene in a DVD will show macroblocks. You want to keep those flaws in there ? Do you think the director wanted to look like that? Just go to the same BD source (if available), and do you think those flaws exists (to the same degree)? The problem is MPEG2 and DVD bitrates. Sure, in motion, those issues are less visible, but I would argue any perfectionist wouldn't want that. I guess it's your choice

Certainly several of these flaws due to the format are correctable, but isn't it true that each change/filter comes at a price? Deblocking and denoising mean blurring, which means less sharp video. With the proper tools, these effects can be diminished, but never absolutely left untouched. In a nutshell, I suppose its solving one issue while creating another, with generational loss following.

I suppose my personal opinion has always been to never spend more time editing than watching, in that if I am going to watch a video once that's an hour long, I shouldn't spend more than an hour's time editing. Life is too short

[poisondeathray]

Certainly several of these flaws due to the format are correctable, but isn't it true that each change/filter comes at a price? Deblocking and denoising mean blurring, which means less sharp video. With the proper tools, these effects can be diminished, but never absolutely left untouched. In a nutshell, I suppose its solving one issue while creating another, with generational loss following.

I suppose my personal opinion has always been to never spend more time editing than watching, in that if I am going to watch a video once that's an hour long, I shouldn't spend more than an hour's time editing. Life is too short

Yes I agree for the most part. I guess it's about finding a compromise or living with the lesser of flaws. But certain releases have notable problems (not just macroblocking) that are clearly better off being fixed.

But I'll tell you right now, that fiddling with the deblocking values for the encoder is the wrong way to do it, especially if source has problems to begin with. Ever heard of the saying "fix the underlying problem" ? You just propogate errors when encoding.

And for the "clean" BD source scenario, you're simply not going to be able to reduce the filesize by 5x, if you're a perfectionist. You can shift values with higher AQ (more bitrate gets allocated to dark areas, flat areas, but this takes away from edge quality primarily), but simply put - you need bitrate to preserve grain and dither. Fine grain and dither /noise is the enemy of compression , but it will reduce posterization and banding . It's one of the majors reasons it's added to BD's by studios in the first place (unlike DVD's , BD's have adequate bitrate and compression to retain the grain. Many DVD's have splotchy grain, or uneven grain from poor I,B,P compression )

[premiumcapture]

Certainly several of these flaws due to the format are correctable, but isn't it true that each change/filter comes at a price? Deblocking and denoising mean blurring, which means less sharp video. With the proper tools, these effects can be diminished, but never absolutely left untouched. In a nutshell, I suppose its solving one issue while creating another, with generational loss following.

I suppose my personal opinion has always been to never spend more time editing than watching, in that if I am going to watch a video once that's an hour long, I shouldn't spend more than an hour's time editing. Life is too short

Yes I agree for the most part. I guess it's about finding a compromise or living with the lesser of flaws. But certain releases have notable problems (not just macroblocking) that are clearly better off being fixed.

But I'll tell you right now, that fiddling with the deblocking values for the encoder is the wrong way to do it, especially if source has problems to begin with. Ever heard of the saying "fix the underlying problem" ? You just propogate errors when encoding.

And for the "clean" BD source scenario, you're simply not going to be able to reduce the filesize by 5x, if you're a perfectionist. You can shift values with higher AQ (more bitrate gets allocated to dark areas, flat areas, but this takes away from edge quality primarily), but simply put - you need bitrate to preserve grain and dither. Fine grain and dither /noise is the enemy of compression , but it will reduce posterization and banding . It's one of the majors reasons it's added to BD's by studios in the first place (unlike DVD's , BD's have adequate bitrate and compression to retain the grain. Many DVD's have splotchy grain, or uneven grain from poor I,B,P compression )

LordSmurf has a pretty good recipe for DVD encoding for GOP stuff, and its shocking sometimes how big companies can release crap and not know how to properly tinker with MPEG-2. I have produced far better quality DVDsI recorded myself on a prosumer camcorder with TMPGEnc than some commercial releases.

One of my guilty pleasures is the 1999 movie Wing Commander, which is easily one of the worst Blu-Ray transfers I have ever seen. Too soft in many places, sound all over the place, and it doesn't help that the movie is pretty bad.

On the camcorder note, my Panasonic X920 recorders in 1080p60 and uses AVCHD but does not suffer from the aforementioned effects, most likely because of the 30ish mbps bit rate and pulling the source directly from the optics.

Anyways, I got terribly off track from my original post but I think I have a pretty good handle on what those values are and know with some certainty that ultimately, even using x264 for a VC1, H.264, or MPEG-2 source, there is going to be some loss of quality, bit rate will ultimately decide how much that is, and using -1,-1 for film sources is probably for the best.