Removing EIA-608 Closed Captions from H.264 without reencode

Question

I'm looking to remove the closed captions (EIA-608) from a H.264 video (contained as mkv) without reencoding.

The closest I've got is using ffmpeg:

    ffmpeg -f lavfi -i movie=input.mkv[out+subcc] -map 0:0 output.mkv

In order to separate the video into rawvideo and subrip components and export out the rawvideo. However this results in a file close to 200GB, which isn't really a sustainable solution.

An ffmpeg based solution would be preferable, but I'm fine using whatever software is necessary.

Answer 1

This is actually possible using bitstream filters. As far as I know I discovered this myself, since everywhere I have looked this is supposed to be unsupported.

The first thing to understand is that for EIA-608 and similar closed captioning standards, the captions are embedded directly in the video bitstream as user data. H.264 bitstreams are stored as a sequence of NAL (network abstraction layer) units. Each unit has a type; user data is stored in a NAL unit of the supplemental enhancement information (SEI) type.

It turns out that ffmpeg has a bitstream filter called filter_units, which allows you to pass or reject NAL units by type. So we can use this to remove all the SEI NAL units, which strips out the captions.

The filter documentation for filter_units says that we have to specify the types by number. According to the latest H.264 spec (Table 7-1), SEI units have type 6.

So the following command will remove embedded closed captions:

ffmpeg -i input.mkv -codec copy -bsf:v "filter_units=remove_types=6" output.mkv

This has worked for me on several files without any problems or side effects.

Answer 2

I came across this post while attempting to remove closed captions from an MPEG-2 video (from an NTSC DVD). filter_units=remove_types=6 didn't work and actually caused a thick black horizontal bar to appear, presumably because crucial video information was removed since the codecs differ. After some experimentation, I discovered that remove_types=178 worked.

Thought I'd post here for anyone else trying to achieve the same in the future. I would've just commented on the answer but I don't have the reputation necessary to do so.

Answer 3

For anyone looking to do this with h265 remove_types=39

It's important to note that for AVC/HEVC, SEI is also used for other information like HDR10+, so using this will result in that metadata being removed, there is an existing open ticket regarding this issue. So if you don't want to lose HDR then it would be best not to remove closed caption.

https://trac.ffmpeg.org/wiki/HowToExtractAndRemoveClosedCaptions