WINRAR no longer free

fwor

Yes, I'm making some assumptions that may or may not be right.

But that doesn't alter the fact that a claim that originally appeared to be impressive is not impressive at all - or is simply not correct.

shadowdragon

well i have just had a trawl on the interwebs and came across a site with data compression tests (this is by no means definative as im not going to spend hours researching it) its quite interesting looking at the table and the different formats as to what compression is acheivable. And if some of those could be compressed to a 98% ratio (which only one or two of the tested files were) that would be a nice compression algorithm even for a standard .doc file.

Even the jpgs that you claim are compressed already were able to be compressed even more (by 25% at the high end stuff), I belive its down to the compression software you use as to its effectivness and if the high end stuff can get 25% theres no reason why it couldnt go higher with new developments.

I am glad winrar comes out on middle ground, as thats what I use.

http://www.maximumcompression.com/index.html

Fifer

Any program claiming to compress any file by 98% without data loss would have to be able to do so with a file it had previously compressed. Patent nonsense.

superscaper

Fifer wrote: »

Any program claiming to compress any file by 98% without data loss would have to be able to do so with a file it had previously compressed. Patent nonsense.

As fwor's random example demonstrates it's as much to do with the amount of information (as in definition in information theory) in the file to be compressed. In fact it'd be very easy to create a file (with very low information) a gigabyte in size and manually compress it 99.99999% losslessly. (Just think of a file of nothing but 1s, lots of data but low information, in fact this very post now contains the compressed lossless version of that theoretical gigabyte file). It's information rate that's the important factor and whether that can be compressed losslessly more or less than 98% is absolutely theoretically possible depending on that value for information.

epz_2

if i recall there was a program that could achieve enormous music compression by taking a different compression approach.

rather than developing pattern matching algorithms it analysed the track and match each frame of music to local database and then creating an index.

The index is then transferred and the pc used to play back the music would have a few hundred megs of local files which it would use to rebuild the song using the index.



its kind of like giving someone an index number to look a book up in the library than trying to compress the book. I suppose you could do the same with any file as long as you had a big enough local database to rebuild the file.

fwor

shadowdragon wrote: »

Even the jpgs that you claim are compressed already were able to be compressed even more (by 25% at the high end stuff), I belive its down to the compression software

JPEG compression is generally not lossless. If you are prepared to lose quality then you will typically be able to compress a JPEG file further.

Any compression algorithm is a trade-off between the amount of compression achieved and the amount of time it takes, but there are diminishing returns, irrespective of the algorithm used. At some point you will have removed all of the redundancy in the file and it cannot be compressed further.

epz: I'd never seen that before, but it's still a trade-off - you still have to transfer the (large) local database to get the information from one place to another.

superscaper

epz wrote: »

if i recall there was a program that could achieve enormous music compression by taking a different compression approach.

rather than developing pattern matching algorithms it analysed the track and match each frame of music to local database and then creating an index.

The index is then transferred and the pc used to play back the music would have a few hundred megs of local files which it would use to rebuild the song using the index.



its kind of like giving someone an index number to look a book up in the library than trying to compress the book. I suppose you could do the same with any file as long as you had a big enough local database to rebuild the file.

I think that's kind of similar to how text to speech convertors work. They don't have a database of every possible word. They just store every possible
phoneme to contruct the words and sentences. I'm not a linguist so my figures may not be accurate but something like well under 100 phonemes compared to 2 million words and infinite number of sentences (which can be created from those phonemes) is a good demonstration of that principle. I suppose it could work theoretically the other way around to compress existing speech/recordings, just the speech equivalent of your music example.

superscaper

fwor wrote: »

Any compression algorithm is a trade-off between the amount of compression achieved and the amount of time it takes, but there are diminishing returns, irrespective of the algorithm used. At some point you will have removed all of the redundancy in the file and it cannot be compressed further.

As an aside (if this weren't off topic enough ), depending on the file we were talking about there are some interesting natural algorithms such as fractal compression that can actually (although not perfectly) go beyond even the redundancy and decompress to an apparently quite lossless copy. Especially works well with anything "natural" such as photos of forests and trees etc because nature itself follows relatively predictive patterns such as fractals and golden ratios.

shadowdragon

superscaper wrote: »

As an aside (if this weren't off topic enough ), depending on the file we were talking about there are some interesting natural algorithms such as fractal compression that can actually (although not perfectly) go beyond even the redundancy and decompress to an apparently quite lossless copy. Especially works well with anything "natural" such as photos of forests and trees etc because nature itself follows relatively predictive patterns such as fractals and golden ratios.

This is the stuff I love reading on here /clap

fwor

I'd forgotten about fractals. Can't you use the theory behind them to prove that the coastline of any country is infinitely long? Perhaps I'm remembering that bit wrong....

As you say, good for some things but not others - I guess they don't deal with straight lines very well.