[FFmpeg-devel] [PATCH] avfilter: add unpremultiply filter

[Paul B Mahol]

On 8/1/17, Tobias Rapp <t.rapp at noa-archive.com> wrote:
> On 01.08.2017 15:31, Paul B Mahol wrote:
>> On 8/1/17, Tobias Rapp <t.rapp at noa-archive.com> wrote:
>>> On 01.08.2017 13:03, Paul B Mahol wrote:
>>>> Signed-off-by: Paul B Mahol <onemda at gmail.com>
>>>> ---
>>>>  doc/filters.texi             |  13 ++
>>>>  libavfilter/Makefile         |   1 +
>>>>  libavfilter/allfilters.c     |   1 +
>>>>  libavfilter/vf_premultiply.c | 307
>>>> ++++++++++++++++++++++++++++++++++++-------
>>>>  4 files changed, 277 insertions(+), 45 deletions(-)
>>>>
>>>> diff --git a/doc/filters.texi b/doc/filters.texi
>>>> index 4089135..a50696a 100644
>>>> --- a/doc/filters.texi
>>>> +++ b/doc/filters.texi
>>>> @@ -14532,6 +14532,19 @@ ffmpeg -i INPUT -vf trim=duration=1
>>>>
>>>>  @end itemize
>>>>
>>>> + at section unpremultiply
>>>> +Apply alpha unpremultiply effect to input video stream using first
>>>> plane
>>>> +of second stream as alpha.
>>>> +
>>>> +Both streams must have same dimensions and same pixel format.
>>>> +
>>>> +The filter accepts the following option:
>>>> +
>>>> + at table @option
>>>> + at item planes
>>>> +Set which planes will be processed, unprocessed planes will be copied.
>>>> +By default value 0xf, all planes will be processed.
>>>> + at end table
>>>
>>> IMHO using a flags-like string "planes=rgb" would be more user-friendly
>>> than a bitmask. At least the documentation should tell which bit refers
>>> to what channel.
>>
>> It is directly related to pixel format.
>
> I'm just wondering whether I can apply the logic of
> AVPixFmtDescriptor.comp to the planes bitmask or not:
>
>      /**
>       * Parameters that describe how pixels are packed.
>       * If the format has 1 or 2 components, then luma is 0.
>       * If the format has 3 or 4 components:
>       *   if the RGB flag is set then 0 is red, 1 is green and 2 is blue;
>       *   otherwise 0 is luma, 1 is chroma-U and 2 is chroma-V.
>       *
>       * If present, the Alpha channel is always the last component.
>       */

I mainly did it bitmask way because of additional code needed to handle cases
when there is no r/g/b but y/u/v planes and user supplied r/g/b only.