Timeline for What simple, effective techniques for obfuscating points are available?
Current License: CC BY-SA 3.0
18 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 9, 2012 at 17:48 | answer | added | Reid | timeline score: 0 | |
| Jun 21, 2012 at 2:57 | comment | added | Llaves | at the risk of belaboring the point, my comment asks how accurate the obfuscated point must. Your reference to "one kilometer" states "it is not possible to derive the original point within (say) a kilometer or so". This is an entirely different concept. | |
| Jun 20, 2012 at 21:15 | comment | added | zebediah49 | I may have phrased that badly. I'm saying that at some level if you can use the information to, for example, determine how close the user is to a point, that can be used to locate the user. The earth is small enough that you need to discard data--most likely by losing enough accuracy that it's not a concern any more. Your other option is to encode it such that it cannot be used for anything useful ("within area X" is no longer an option). | |
| Jun 20, 2012 at 21:03 | comment | added | Reid | @zebedia, I don't understand your issue. The point is that a user might input the locations of their home, and I don't want to associate that location with their other metadata or even with the fact that someone at a particular location participated in the site. How is your approach relevant to that? | |
| Jun 20, 2012 at 20:29 | comment | added | zebediah49 | One problem with that kind of accuracy: If you can determine if it's within "a KM" of a point, I only have to brute-force 500M points to cover the surface of the earth. That's totally easy. Since you can't do better than that, you might as well just round to the nearest minute of lat/long, and store that. | |
| Jun 20, 2012 at 19:59 | answer | added | MerseyViking | timeline score: 1 | |
| Jun 20, 2012 at 17:30 | comment | added | Reid | @Matthew, that sort of analysis is feasible, yes, but we don't need to join the points with points in another dataset (I envision questions more like "how many points in this polygon from another dataset). | |
| Jun 20, 2012 at 17:27 | comment | added | Reid | @Mersey, I don't need to recreate the original point. But it does need to be a geographic point that I can use in later analysis. (The points are citizen science observations, but we may be able to get away with a reduction in precision that comes with obfuscation.) | |
| Jun 20, 2012 at 17:24 | comment | added | Reid | @Llaves, I do actually: "within a kilometer or so". But I would hope the obfuscation level is a parameter to the algorithm. Regarding your second comment, yes, the metadata allows association of points (e.g., one user might enter the same point many times). And an algorithm that results in the same obfuscated point given the same original point is fine; but if the algorithm doesn't do that, I can't recover the original point (that's the whole reason for the question) in order to test if the same obfuscated point should be used. | |
| Jun 20, 2012 at 17:23 | history | edited | Reid | CC BY-SA 3.0 |
rm request for tag since someone did it, thanks!
|
| Jun 20, 2012 at 13:34 | comment | added | Matthew Snape | Will the points be compared with other users data/other datasets as a part of the service? | |
| Jun 20, 2012 at 9:54 | answer | added | user173 | timeline score: 7 | |
| Jun 20, 2012 at 8:56 | comment | added | MerseyViking | Do you need to be able to recreate the actual location from the hashed data, or will it just be used for confirming a person is where they say they are? If it's the latter, a one-way hash, hashing a salt + the WKT of the geometry would suffice. If it's the former, then you'll have to have some function somewhere to do the inverse transformation of your hash function - a two-way hash. | |
| Jun 20, 2012 at 7:41 | history | edited | Ian Turton |
edited tags
|
|
| Jun 20, 2012 at 4:29 | comment | added | Llaves | A second question: you mention metadata and being able to reconstruct the true point if the entire database is compromised. If the metadata doesn't allow you to identify obfuscated points associated with the same "true point", then how can someone reconstruct the "true point" from repeated random samples if you can't associate them with each other? On the other hand, if the metadata does allow you to associate the points, then when you are asked to report again the location of some already obfuscated point, just return the same obfuscated value returned all the previous times. | |
| Jun 20, 2012 at 4:14 | comment | added | Llaves | Your requirements don't mention what sort of accuracy you wish to maintain, you only focus on the obfuscation requirement. The following algorithm trivially satisfies the requirements you listed, but is rather worthless: map each point to 0° N, 0° east. Presumably you also want to satisfy some criterion, like the obfuscated point is within x km of the actual point. | |
| Jun 19, 2012 at 23:51 | answer | added | AHigh | timeline score: 0 | |
| Jun 19, 2012 at 22:13 | history | asked | Reid | CC BY-SA 3.0 |