[x264-devel] Pointers for experimenting with foveated compression

[Luke Hsiao]

Hi x264 developers,

My labmates and I are interested in exploring some modifications to x264
in the context of foveated video encoding. That is, experimenting with
techniques for achieving better compression by discarding additional
information in a viewer's periphery. For example, dropping color, or
zeroing out residuals for the macroblocks in the peripheral vision.

Previous work used the quant_offsets in the properties of an x264
picture to change QP in the periphery, which is relatively
straightforward to implement. For some of the other things we'd like to
try (e.g., dropping chroma, zeroing out residuals) it's less obvious
what the best way to go about implementing it is.

Could anyone give us some pointers on where we might look or how we
should approach these modifications?


--
Luke