Method Review: Does It Make Sense?
Hello Mrayness,
I visited your link but didn't want to subscribe and only read the abstract of Hamza and Krim's paper. Given that f(x) the Geodesic Shape Function, is the sum of absolute distances between centroid x and all other centroids in the set, it seems logical that the same x occurs in the Shape Distribution.
I'm not familiar with the Jensen-Shannon method for comparing paired kernel densities. However, if the dimension of each p(x) is indeed a probability, they can be compared provided that the number of degrees of freedom for each p(x) is taken into account.
The median absolute deviation is not a robust statistic so I wonder whether the mean of absolute differences, and the mean of the sum of squared differences, between centroid x and all other centroids in the set would more effectively define the Geodesic Shape Function.
Did you check out the Eclipse Modeling Framework? My son loves it but his opinion is necessarily biased!
Kind regards,
Jan W Merks ■
Untitled
Many thanks Sir.. 
■
Method Review: does it make sense?!!
I implemented an algorithm for representing and matching 3D objects using statistical methods, using Java programming language.
As I am a programmer, I need some statistician to review me the statistics models used:
The whole paper is found:
www.springerlink.com/index/0L48BVDX0B86NQVE.pdf
The algorithm goes like this:
1. Object Representation:
convert the object from 3D to 1D vector X={f(x1), f(x2), ...., f(xn)}
where n: number of centroids (triangles) in the mode,
f(x) is the geodesic shape function (summation of geodesic distance from centroid x to every other centroid)
2. Shape Distribution:
to describe the structure of the data represented by vector X, use kernel density:
p(x) = 1/nh summation(K((x-Xi)/h))
where h: bandwidth, K: Guassian kernel
I need to make sure of :
1- the parameter h: I used: h= 1.06 * power(n,-1/5) * median absolute deviation
(this is different from the one used in the paper)
2- In the equation above, what is x?
3-For conparing two kernel densities, Jensen-Shanon method is used, do both vectors X1 and X2 need to be of the same size in order for the comparison of thier kernel densities be correct?
Regards; ■