Adaptive Matrix
 

has someone been able to work with the adaptive matrix feature of HCenc (from ver 0.20). It needs a bit of programming and I am not upto it.

I don't, but just FYI : from 0.21 there is a built-in feature that deals automatically with the matrix according to the luma in the frame. I don't understand how and why, but look at the *LUMGAIN parameter in the doc.

I read that but dont know how the output will be if I used it, although theoritically once HC detects dark scenes it will raise the br accordingly, which sounds good.

I'm in the same sitaution as you are : I need to test that to understand really what is behind it :).

Here's some insight by hank himself.

Might help in reducing the blocks in dark scenes. It seems similar to AutoQmatenc "Dark mode", i.e. it first detects if the encode required os for low lit video segment (checks the luminance) and then engages the dark mode automatically.

Hi all,

I've been playing with adaptive matrix dll programming a year ago, and if you're interesting in, I have a c++ source that replicates the fortran sample.

It's a visual studio 2005 project, but I think it can be compile with others C compilers.

Fabrice

What have you implemented in your program, can you describe how have you done the control on the matrix being used.

Hi,

I was playing with motion (factor a, limited to 2) and luminance (b factor, with value between 0.9 and 1.05) and increasing the notch matrix by those 2 factors. I increased less the intra matrix than the inter matrix.

The 'algorithm' was:
- inter: notch_inter[i] * (1+ (a* row * column / 64)) * b
- intra: notch_intra[i] * (0.9+a/5) * b

But you can play as you want, with the factors you want. As I didn't get a 'wow' result (a lot smaller file with the same visible quality) , I'm still using the standard notch matrix...

Fabrice