User talk:Ste4k/usit

We could use the algorithms here. Ste4k 14:20, 26 July 2006 (UTC)


 * -4 Octet CRC32-|--4 Octet Image Length-|---4 Octet Data Sampling---|
 * xxxxxx xx|xxxx xxxx|xx xxxxxx|xxxxxx xx|xxxx xxxx|xx xxxxxx|xxxxxx xx|xxxx xxxx|xx xxxxxx|xxxxxx xx|xxxx xxxx|xx xxxxxx|
 * ---0--|---1---|---2---|---3--|---4--|---5---|---6---|---7--|---8--|---9---|---10--|--11--|--12--|---13--|---14--|--15--|

we need a real image for this. Ste4k 15:14, 26 July 2006 (UTC)

sampling alg
This is only a starter, we need to throw some sampling tests at it. Ste4k 16:34, 26 July 2006 (UTC)

refactor the USIT
...section up to the first section after the introduction, before the encoding section. Looks like a good start, but you probably need to better define your target audience and purpose of the 'proto-article'.

Why not use CRC64? If I remember the theory, taking alternate even and odd sequential samples and combining into 64 bytes will hash out as a unique longer CRC, since you aren't combining adjacent data in the individual summations. I did such back in '90 to get 'the novel' and rare CRC32's of that era.  Well, it worked for us... a measly two million line of code project. :-)

Good job! // Fra nkB 00:19, 27 July 2006 (UTC)

Feel free to edit Fabartus. We are still tossing this around over on this side. This was today's draft and combination after discussion on IRC. About CRC64, I haven't that at my availability. I suspect, though, that the data for the table would take quite a bit. In anycase, though, we end up with a 96-bit id code that is unique per size, and carries a 4 byte signature that is dependant on the file. Once we have a normalized and tested random routine the resulting "odds" of a collision will end up being the worse with quite fewer coding problems, much better speed, etc. There are some other problems though that neither CRC nor this proposed USIT currently address. Depending on the application, one could theoretically simply modify one pixel in a photo (a small red dot in the bottom right extreme corner) and a human would never notice that it was a different image. The algorithm would notice, though and declare that to be a "different" image. So if the application is to find duplicates, it will work just fine. But if someone wishes to defeat it (i.e. go around a cpvio) then it will not detect such an easy defeat. Ste4k 06:34, 27 July 2006 (UTC)