Introducing Daala part 4: Chroma from Luma (xiph video codec) | Hacker News Reader

Introducing Daala part 4: Chroma from Luma (xiph video codec)

people.xiph.org

53 pointsjobstijl8 comments

  xiphmont_: huh demo4:CfL is on HN already
  jobstijl0: I posted it, didn't saw you where still working
             on it
  xiphmont:  ah, it wasn't actually due to be up for real 
             until I got back from vacation in two weeks
  xiphmont:  no harm, just 'agh, nothing is tested yet!'

I noticed

> [some RDO graphs will be nice-- add these]

Thank you Monty! Awesome as always and already waiting for the next part - no pressure :)

And just because I know someone is going to ask: Motion prediction in H.264 is based on luma, but there's also a chroma option. I've done some quick searches and found a patent for HEVC that covers using the correlation for motion prediction.

"Luma predicts from luma, and chroma predicts from chroma. There's little or no attempt to account for remaining cross-channel correlation." Interesting, would have thought these are one of the first things to try...

They did try in HEVC! It was scrapped because it decreased the encoder performance and its benefits didn't show up in their primary test metric (PSNR).

Great tutorial.

They use some terms like DC(or TF) without explaining what they mean. I could not understand them without significant effort.

For example, for "DC" I know(after careful analysis) that they mean constant average value, it could mean Discrete Cosine, or Discrete Component. What does it mean? "Direct current"?

How Direct current will make any sense in a digital signal article?

DC is the low frequency (zero frequency) component of the signal and is indeed obtained by simple averaging.

The term DC is borrowed from electrical engineering. https://en.wikipedia.org/wiki/DC_bias explains:

    The term originated in electronics, where it refers to a
    direct current voltage, but the concept has been extended
    to any representation of a waveform. The term's use is
    extended to two-dimensional transformations like the
    discrete cosine transform used in JPEG.

TF is defined in the previous demo page, http://people.xiph.org/~xiphmont/demo/daala/demo3.shtml. It means Time/Frequency Resolution Switching, a idea from Xiph's Opus codec.

Reply on news.ycombinator.com