X264 encoding Ringing around subtitles

[MatLz]

Hello everyone. In X264, is there a parameter to adjust to limit ringing artifacts around subtitles?

[Inspector.Gadget]

Why not use softsubs?

[wyti]

you can try to disable Adaptative quantization and psy-rdo (--aq-mode 0 --psy-rd 0:0) to help preventing them, but your whole video may look worse because of desactiving thoses options, the best way (if possible) is to use softsub with external subtitle files.

[LoRd_MuldeR]

Before you completely disable AQ and/or Psy-RDO and loose powerful features of x264, you may want to try lowering them a bit!

BTW: Did you use Psy-Trellis? If so, you should be really careful with strength here...

[MatLz]

In fact I recode animes with subtitles incrusted (euh...I'm french...Must I say "hardsubbed" ?)
I always use high parameters like tesa, subme 9-10, always trellis....so yes I can try without aqmode and psyrdo. Thanks

[LoRd_MuldeR]

In fact I recode animes with subtitles incrusted (euh...I'm french...Must I say "hardsubbed" ?)
I always use high parameters like tesa, subme 9-10, always trellis....so yes I can try without aqmode and psyrdo. Thanks

Use UMH and save a lot of time! Also try lowering the strength of AQ/PsyRDO before you completely disable ;)

BTW: Where is the hardsubbed content from?

[Dark Shikari]

Psy-RD likely has nothing to do with it.

Lowering AQ strength will improve the subtitles, but at the cost of everything else.

[MatLz]

In fact the artifacts are around the subtitles in the bottom and around the japanese symboles in the introduction...so when there is a black and white contraste.
And about quant matrices? Can it do something to that? (Sorry if I'm slow...I use a phone to post)

[Dark Shikari]

In fact the artifacts are around the subtitles in the bottom and around the japanese symboles in the introduction...so when there is a black and white contraste.
And about quant matrices? Can it do something to that? (Sorry if I'm slow...I use a phone to post)Custom quantization matrices aren't generally recommended with anime, but a very lowpass-y matrix will help avoid ringing (but at the cost of blurring).

[LoRd_MuldeR]

And about quant matrices? Can it do something to that? (Sorry if I'm slow...I use a phone to post)

You usually don't need to use a custom quantization matrix.

The things that people tried to achieve with custom matrices in the past are now done by x264's AQ and Psy RDO/Trellis - only in a better and more general way ;)

Also I think it was said that x264's Psy-opts are tuned for the "flat" (default) matrix and may not perform that well with other (extreme) matrices...

[poisondeathray]

A higher bitrate usually helps too :)

What are your source specs? dimensions? and what did you use for bitrate?

[MatLz]

I downscale 1280x720 perfect sources to 1024x576. Just to limit size. The resulting bits/pixel.frame is higher than the source. I will do Dark_Shikari and Lord_Mulder say about aq.....and try try try.

[MatLz]

Custom quantization matrices aren't generally recommended with anime, but a very lowpass-y matrix will help avoid ringing (but at the cost of blurring).

If after that I use negative values for deblock, it will sharpen the blurring result? Or deblock is before quant matrices?... How is the order chain?

[LoRd_MuldeR]

Quantization happens when a frame is compressed. The deblocking is applied after the frame has been decompressed.

Also negative numbers for the Loop-Deblocking won't "sharpen", they will just deblock/smooth less aggressive and hence preserve more detail/sharpness (at the risk of more blocking).

It's seems that negative Deblocking numbers like -1:-1 are helpful for "Film" material, while "Anime" likes positive values like +1:+1. Compare "--tune film" vs. "--tune animation" ;)

[MatLz]

Thanks a lot for all your explanations. I will do tests and tests and give the results in this thread of course. And thanks for your softwares too.

[nakTT]

Custom quantization matrices aren't generally recommended with anime, but a very lowpass-y matrix will help avoid ringing (but at the cost of blurring).
May I know how can I set this lowpass-y matrix? I planned to backup my animation video in a couple of week.

:thanks:

[MatLz]

Hi nakTT, I think "Y" is Luma, so we need to find a lowpass-luma matrix.
--aq-mode 0 seems to do good results but I will try more tests with lower values of --aq-strength with/without --psy-rd.

[Dark Shikari]

Hi nakTT, I think "Y" is LumaNo, it's the part of the word that makes a verb into an adjective ;)

[MatLz]

No, it's the part of the word that makes a verb into an adjective ;)

In that case of technical discussion it's chinese for the poor multilingal I am! :'(
I found some quant matrix but I don't know if they are lowpass-y or not!

[Dark Shikari]

In that case of technical discussion it's chinese for the poor multilingal I am! :'(
I found some quant matrix but I don't know if they are lowpass-y or not!Something with much bigger numbers in the bottom, right, and bottom-right than the top-left is a lowpassy matrix.

[MatLz]

Ok, thanks Dark Shikari. It speaks to me a little more! However, are these "custom" matrix compatible with multimedia hard disk? (I have bad memories with custom matrix for Xvid not compatibles with some dvd players)

[nm]

Ok, thanks Dark Shikari. It speaks to me a little more! However, are these "custom" matrix compatible with multimedia hard disk?If you mean a mobile playback device, probably not unless it supports high profile. Most standalone boxes that play HD H.264 video do support HP and therefore CQMs though.

[MatLz]

Thanks nm, good to know.
So... for the ringing artifacts around subtitles, after few tests....the winner is.....?......are --aq-mode 0 --psy-rd 0.00:0.00
I have try CQM lowpass to but without real effects.