Unblock - statistical deblocking filter (blind)

[Fizick]

I was need to deblock photo slides copy and discover interesting debloking program at:

http://www.assassinationscience.com/johncostella/unblock/

From the original paper:

We don't have the original image, so we can't be sure that there isn't a sharp discontinuity in the original image that happens to lie right on the boundary. If we removed all discrepancies unconditionally, we would automatically blur out any such sharp edge that happens to unfortunately fall in the wrong place. It is impossible to avoid such an error in every possible case, but we can do so statistically. For any given image, the distribution of the magnitudes of the block boundary discontinuity errors will be determined by the amount of compression employed (the "quality factor"). By computing the distribution of boundary discrepancies across the entire image, and comparing it to the distribtuion of "discrepancies" calculated in the interior of the blocks, we can quantify, statistically, the amount to which the compression has introduced block artifacts. Now, what are we to do with these statistical distributions? A reasonable ansatz is that we should reduce each boundary discrepancy by an amount that will make the interior and boundary discrepancy distributions the same, as far as possible.

it is first part of my message - i need in thread number for plugin doc :)

[Soulhunter]

If Pegasus would interpolate the chroma like Unblock does, it would look much better! >.>


Merry X-Mas n' Bye

[Fizick]

Plugin and doc is almost ready. But I try contact author for some question about chroma deblocking.

I forget that X-mas is today :)
In our country it is 7 january, but new year is 1 january too :)

[redfordxx]

Plugin and doc is almost ready. But I try contact author for some question about chroma deblocking.
But it is only needed when you go away from YV12 --- otherwise is the chroma treated same as luma, right?
I forget that X-mas is today :)
In our country it is 7 january, but new year is 1 january too :)Yeah, as one friendof mine (came from somewhere around Vladivostok) told me, she starts shopping for "her" Xmas only after "our" Xmas, when there are the greatest sale prices;-)

Merry Xmas to all...

[redfordxx]

I read the paper. I really wonder how it is possible, that it works with lines treated independently...

Moreover, I am little unsure in the part 3.5 of the paper...maybe my view is then wrong and needs correction:

Is your goal JPEG or MPEG or both?

In case of MPEG I think it could be interesting to incorporate weighting by quantizers (two adjacent blocks with different quantizer can differently contribute to the discrepancy). Interior discrepancy values will also depend on quantizer (higher q=smaller interior discrepancy, bigger boundary discrepancy).

Also, error magnitude can be estimated using quantizer and matrix values.

[Fizick]

redfordxx,
no, weighting by quantizers will be incorporated in your filter, not in the Unblock :)
it is blind like blindPP function.
(it is not point what is better. There are many methods. Unblock is Unblock as is)

[Soulhunter]

But it is only needed when you go away from YV12...
Chroma deblocking? No! Chromaupsampling? Yes! Just wanted to point out that the comparison on this page could give a false impression... Because the Unblock results have upsampled chroma but the Pegasus ones have not... The pointresized chroma of the Pegasus results, together with the global 2x pointresize, gives the impression that the Pegasus results look much more blocky than they really are... As this comparison is about deblocking and not about [relative easy to accomplish] chroma chromaupsampling, let us compare just the deblocking... (http://img508.imageshack.us/img508/6259/comp1bq2.png) ^^


Bye

[Fizick]

Released UnBlock version 1.1