Home > Archive > Compression > September 2006 > Lossless re-compression
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
Lossless re-compression
|
|
| Stefan Monnier 2006-09-19, 6:55 pm |
| Can anyone point me to information about lossless re-compression,
i.e. compression using a lossless codec like jpeg but of a stream known to
come out of the jpeg decoder and hence can be reencoded to the
original bitstream.
Of course, it becomes more interesting when some part of the image has been
changed, in which case we'd like to make sure that only the part that has
been changed may lose info upon re-compression.
Stefan
| |
| niels.froehling@seies.de 2006-09-20, 6:55 pm |
| > Can anyone point me to information about lossless re-compression,
> i.e. compression using a lossless codec like jpeg but of a stream known to
> come out of the jpeg decoder and hence can be reencoded to the
> original bitstream.
>
> Of course, it becomes more interesting when some part of the image has been
> changed, in which case we'd like to make sure that only the part that has
> been changed may lose info upon re-compression.
Possibly any predictive image compressor with quantizer-feedback is
deterministic (re-codable) - if you know the exact setup of parameters
that effect the prediction and quantization.
A lossy variant of CALIC would be re-codable (if with quantizer
feedback) - if BMF or JPEG-LS are really re-codable I didn't determine
(they have a 'near-lossless' mode that maybe isn't a feedback).
BTPC/APT and the PyramidWorkshop are all re-codable - and store these
parameters too.
Because of the progressive nature of the BTPC algorithm, the change of
a single pixel between decoding and recoding, may change surrounded
regions (dependent of the position only a direct neighbour or in the
extreme case upto a region of (1 << progressive levels) pixels in
square).
Re-codablility in that kind of feedback is a natural property, the
encoder only sees allready quantized information (it has no knowledge
of the original state) insofar encoder/decoder losslessy code a
pre-processed image - the loss happens as a (quasi) independent
pre-process, if the quantizer can be iterated over though-put without
changing it any more after the first iteration.
BTPC/APT: http://www.intuac.com/userport/john/apt/
PyramidWorkshop: http://sourceforge.net/projects/pyramidworkshop
> Stefan
Niels
| |
| Stefan Monnier 2006-09-26, 6:55 pm |
| >> Can anyone point me to information about lossless re-compression,
[color=darkred]
> Possibly any predictive image compressor with quantizer-feedback is
> deterministic (re-codable) - if you know the exact setup of parameters
> that effect the prediction and quantization.
I was thinking about the situation where the exact parameters of the
original compression are not known. I.e. you know there exists a compressed
form that is lossless, but you have to search for it. Maybe in order to
rule out trivial answers, you may need some hints as to the target size
(otherwise you may be able to encode losslessly by not quantizing at all,
even in jpeg ;-).
Stefan
|
|
|
|
|