Re:Knusperli: A deblocking JPEG decoder
by psandersen on Monday March 19, 2018 @01:12AM
(#16616296)
Additionally, I'd love to see a decoder add noise proportional to the uncertainty and surrounding structure as over smoothed images look unnatural as well. See activity masking demo https://people.xiph.org/~jm/daala/pvq_demo/
Pik can synthesize noise. https://github.com/google/pik
yeah, the deblocked image looks a little softer. Makes sense, the optimization objective encourages it. Sometimes the block boundary aligns with an image boundary. Agree that incorporating a DNN could be a cool solution.
Re:Knusperli: A deblocking JPEG decoder
by lokopodium on Sunday March 18, 2018 @10:46PM
(#16615911)
The same should be applicable to MP3s. Also, the continuity measure could include derivatives as well.
Re:Knusperli: A deblocking JPEG decoder
by disposablename on Monday March 19, 2018 @11:28AM
(#16619120)
DNNs are not magic, and not everything needs them, especially a fucking JPEG decoder, come on.
In essence, it computes the uncertainty from the JPEG quantization step (the source of most of JPEG’s compression) and uses that information to select a decoding that minimizes a particular error metric (here, discontinuities at block boundaries).
But it’s not limited to just that, in principle. It could optimize towards any decoded block within the quantization range given a suitable prior - and I imagine that this could be paired with e.g. a DNN to faithfully reconstruct an image from a JPEG with much better accuracy.