I am trying to implement an algorithm that extracts,from a 2d data map,
subsets of data in wich energy density is higher than the mean. If you
represent this data using colours in the desktop , visually you can see
these subsets.
I don't know wich terms to search to get more information. Maybe
algorithms of pattern fitting from image processing, but I think these
are not exactly what I need ( I have not many idea about image
processing). My subsets always can be aproximated by a rectangle or an
ellipse. They are like clouds and I only need an aproximation of wich
area they are in.
There are existing algorithms for this? Wich terms could I search to
get information? The algorithm would need to be efficient.
Thanks.

[ comp.ai is moderated.  To submit, just post and be patient, or if ]
[ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
[ ask your news administrator to fix the problems with your system. ]

juanjo75es@terra.es wrote:

> I am trying to implement an algorithm that extracts,from a 2d data map,
> subsets of data in wich energy density is higher than the mean. If you
> represent this data using colours in the desktop , visually you can see
> these subsets.

What do you mean with 2d data map?

If the index of an array is one dimension and the
values of the elements of the array are the other dimension,
then e.g.

  34 5 44 54 53 356 23 46 3 54 8 3 7 3 237 39 46 29 47 329 24

is a twodimensional data set.


Can you give an example of your data?

If you need all values higher than the mean,
you first have to calculate the mean and then extract
all values that are higher than it.

Isn't this an easy task?
Is highest performance needed, so that you ask for
certain algorithms?
Or can id done "intuitively", as said above:
a) calculate the mean b) select items with value > mean

Maybe for higher performance one can use approximative
algorithms to calculate the mean or so.
(Maybe a newsgroup, specialized on algorithms would be better
here?)



> I don't know wich terms to search to get more information. Maybe
> algorithms of pattern fitting from image processing, but I think these
> are not exactly what I need ( I have not many idea about image
> processing).

Image processing maybe provides some useful algorithms for
you, but you first have to clarify your needs.


> My subsets always can be aproximated by a rectangle or an
> ellipse. They are like clouds and I only need an aproximation of wich
> area they are in.

Yes, this smells like algorithms that are derived from the
area of image processing.
You maybe need some kind of SPATIAL DATA STRUCTURES...
this make sense, if you want to locate certain areas,
e.g. if you want to implement a database, where you can
look for "is <foo> inside <bar>"?

But maybe you only need the (yes, already existing)
algorithms for cluster analysis.

I don't know if both approaches (spatial datastructures and
cluster analysis) already is used together, but if I would
have to solve similar problems I would think about
combining these two aproaches.

It depends on your needs.


> There are existing algorithms for this? Wich terms could I search to
> get information? The algorithm would need to be efficient.

OK, on the first look it seems cluster analysis and
spatial data structures can be useful to you.

For cluster analysis look inside some good mathematical
books, or maybe special books on signal analysis
(you need cluster analysis in electrotechnics/electronics
  but it is also used in sociology for example).
I have only german books in mind which I could recommend you.

If you look for spatial data structures, look for the books from
H. Samet....
wait a moment .... well, here is a link:
   http://www.cs.umd.edu/~hjs/design.html

He is the first address for such datastructures.
Both books are excellent, and there are also
algorithms for those datastructures
in  "The Design and Analysis of Spatial Data Structures".



Ciao,
    Oliver

[ comp.ai is moderated.  To submit, just post and be patient, or if ]
[ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
[ ask your news administrator to fix the problems with your system. ]

I'm not sure I completely understand what you need, but you might try
gaussian distribution or gaussian curve (german mathematician Johann
Gauss).

[ comp.ai is moderated.  To submit, just post and be patient, or if ]
[ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
[ ask your news administrator to fix the problems with your system. ]