[Sls-mhdc] some results in response to action items

[Enrico Magli]

Dear all,

regarding the ESA action items, I have uploaded two xls files on CWE (private folder of May 2017 meeting), with the following results.

- Compression efficiency of FAST predictor. We have run the new simplified initialization on all images in the dataset. The xls file reports the compression efficiency loss with respect to the standard initialization; the results use the range encoder, but I do not expect significant differences employing a different entropy encoder. As you will see, the performance loss is very small; in few occasional cases, the FAST predictor actually outperforms the original one. All in all, my assessment is that the possibility to achieve fast hardware implementation largely outweigh the small performance loss, and in any case the use of FAST would be optional.

- Syntax for signaling of quantization step sizes for rate control.

Option 1: if just one Q step is used every 2^n lines with their spectral channels, the overhead is negligible even for n=0.

Option 2: if one Q step is used for every band every 2^n lines, then the overhead may become nonnegligible. The corresponding xls file exemplifies this, showing results for 4 images, where the Q values are written uncompressed, or differentially compressed with an exp-golomb code of order zero.


- Relative error quantizer. I am finishing to collect results and will send them shortly in a separate email.

Regards,
Enrico