3 ms·
This is interesting in terms of information theory. When you compress information you are basically figuring out the pattern behind something, and encoding that
by Houshalter 13y ago
This is interesting in terms of information theory. When you compress information you are basically figuring out the pattern behind something, and encoding that pattern instead of the entire image. For example, PNG which assumes that pixels are more likely to be the same or similar to the pixels above or to the left of them. So you only have to encode how much difference each pixel has to it's neighbors, rather than the entire value of every pixel.
If you were to randomly corrupt a highly redundant format, like a bitmap, it would just change a few pixels. On more compressed formats like JPEG it seems to affect the entire image, and in very specific ways (mainly the color of every block of pixels after the point that was corrupted.)
If you corrupted a perfectly compressed image, it would give you a different image entirely, possibly of something very similar. I.e. if you had a image format very good at compressing faces, corrupting it would result in a different face entirely, not randomly misplaced pixels or colors. And the face would be similar to the original, maybe with stuff like a different nose type, or an extra freckle.
The corruption is revealing what kinds of assumptions it is making about the content.
- underwater 13y agoThat's similar to how Xerox copiers were switching numbers -- http://www.theregister.co.uk/2013/08/06/xerox_copier_flaw_means_dodgy_numbers_and_dangerous_designs/ http://www.theregister.co.uk/2013/08/06/xerox_copier_flaw_me... Though that was due to overly aggressive compression instead of corruption.
- reginaldo 13y agoNot just Xerox, btw... My HP printer/scanner does the same. After reading that post I started paying attention and eventually saw the effect a few times. I just wonder how many copies with the wrong numbers I've produced so far...
- theon144 13y agoI never thought about it that way, thank you! It makes me wonder what audio format could be fun to corrupt this way. FLAC, maybe?
- driax 13y agoFLAC would probably be a bad example, since it is lossless. It makes fewer assumptions about the content of the audio file, since it anyway has to encode all of the noise and minor stuff. Making too many assumptions of the content would only necessitate a facility to introduce specified noise into the audio later. Which I can't imagine would be both fast and small.
- DebianUser 13y agoI don't think it has much to do with the fact that it is lossless, but rather that it does not compress in a suitable way. In the scheme of things outlined by GGP. Whether such scheme implies lossy is a question that needs to be demonstrated imho.
- jensnockert 13y agoEach block is independent in FLAC (and quite short), and therefore wouldn't generate as interesting artifacts.
- saidajigumi 13y agoHeh, at my first startup I used to get members of the audio codec team wandering over when there was a bug. I'd listen to the corrupted playback, draw out the codec block diagram, and point right to the likely error locations. As another commenter mentions, these were all lossy codecs. The character of noise would be very interesting much of the time -- effectively turning the decoder into a synthesizer.
- dspillett 13y agoMost audio and video formats are based on encoding relatively short block independent of what has come before - otherwise when you skip into the middle of the track the player would need to read everything up to that point before being able to restart play. Many video formats do look back a fair number of frames, so you do sometimes see corruption remain and spread for a few seconds, but then you hit a key frame or block boundary where everything resets and all is suddenly well again. Audio compression is usually a different bag entirely: over a given time period the compression algorithm looks for what it can leave out or merge from the input signal over that time because human ears are unlikely to hear the difference, there is much less opportunity (compared to still images and video) to be able to encode "copy that chunk from a frame or few back, rotate/skew/what-ever it a small amount, then change these few blobs".
- cscheid 13y agoReminds me an argument about crossword puzzles I think comes from Shannon. It goes like this: in a sufficiently "fault-tolerant" language (ie with low per-character information content, which sorta gives large Hamming distance), crosswords become impossible, because puzzles won't be satisfiable: words could never line up right. But in a sufficiently compact language (many bits per character in words, and Hamming distance thus tending towards zero), crosswords are impossible as well, but now because puzzles would be ambiguous: too many words that fit. Somehow, natural languages appear to fit somewhere in the middle.
- cbr 13y agoin a sufficiently compact language crosswords are impossible as well, but now because puzzles would be ambiguous: too many words that fit This ignores the clues, which restrict which words are acceptable in a space. While the clues are mostly there to make the puzzle easier, resolving ambiguity is a secondary purpose.
- cscheid 13y agoSo I looked it up, and it's actually on the seminal 1948 paper: http://cm.bell-labs.com/cm/ms/what/shannonday/shannon1948.pdf http://cm.bell-labs.com/cm/ms/what/shannonday/shannon1948.pd... It turns out that I either hallucinated (incorrectly, perhaps) one side of the argument, or I read it from some other place, because the paper only talks about one side, the one about feasibility. The relevant paragraph follows: "The redundancy of a language is related to the existence of crossword puzzles. If the redundancy is zero any sequence of letters is a reasonable text in the language and any two-dimensional array of letters forms a crossword puzzle. If the redundancy is too high the language imposes too many constraints for large crossword puzzles to be possible. A more detailed analysis shows that if we assume the constraints imposed by the language are of a rather chaotic and random nature, large crossword puzzles are just possible when the redundancy is 50%. If the redundancy is 33%, three-dimensional crossword puzzles should be possible, etc." Anyway, to somewhat rescue a botched story, the ambiguity part reminds me of the mindblowingly awesome and famous puzzle by Jeremiah Ferrell on the eve of the 1996 presidential election, which clue was "Lead story in tomorrow's paper", with possible answers BOBDOLE ELECTED or CLINTON ELECTED, both satisfying all crossclues as well! Amazing: http://cdn0.dailydot.com/uploaded/images/original/2013/2/12/nytimescrossword.jpg http://cdn0.dailydot.com/uploaded/images/original/2013/2/12/...