Page 1 of 1

Robust density computation

Posted: Thu Oct 28, 2010 7:50 am
by Aurelien
To make the density computation more robust, I think we should estimate densities using several points instead of the distance between each point and its nearest neighbour. I propose that we use one of the two following possibilities instead of the one which is currently implemented in CC :
- estimate densities by counting points in a given volume, typically a sphere centered at the point to be treated whose radius is specified by the user
- estimate densities by computing the smallest volume (typically using a sphere) containing some amount of points specified by the user. We can notice that the current implementation of density computation in CC is a special case of this general approach, where the number of points which are considered is 2

Indeed, this would somehow provide a less accurate but more reliable information, by "smoothing" the densities scalar field (like it is done for noise removal in image analysis, using gaussian filters). The user could still use 2 points for this computation, but he/she will have the choice in the way it is handled.

@ Daniel : what do you think about this suggestion ? I'im currently on it, and I can commit modifications after I get your approval.

Re: Robust density computation

Posted: Thu Oct 28, 2010 9:59 pm
by daniel
Indeed!

The main purpose of the actual method was to get a very fast evaluation of density. I was about to implement just this morning the second solution you proposed!

So you can do this with all my blessings (but with a nice dialog ;).