[FFmpeg-devel] rectification filter

[Nicolas George]

Le tridi 23 thermidor, an CCXXII, Michael Niedermayer a écrit :
> thats not practical
> a 1024x768 image would need to be upscaled to 262144x196608 to get
> 8bit precission from a nearest neighbor resampler as basis

I am not sure this makes sense. Scaling is already an approximation, and
without gamma correction, it certainly does not yield 8 bits precision. I
believe in most cases, 2× or 4× upscaling would look plenty good enough for
most cases.

> also see vf_perspective.c which supports bilinear and bicubic
> interpolation, these surely could be shared and exist already.
> 
> of course we could push it with just nearest neighbor and work on that
> later, but i dont think only nearest neighbor and leaving it at that
> is a reasonable choice, its too poor quality wise

Certainly, we want that all filters that need it eventually get
interpolation code that produces nice-looking images, is optimized and
configurable. Preparing the framework for that is a worthy endeavour, and
should be done, as much as possible, before introducing new filters that
require it.

But right now we have a filter that is ready for integration with nearest
interpolation: surely it is better to merge it now rather than having it
wait in someone's private tree while we bikeshed the common interpolation
API. Is it not?

Regards,

-- 
  Nicolas George
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20140810/2b19fdf7/attachment.asc>