[Scilab-users] result of spec function generate an error #144

[Serge Steer]

Le 19/12/2012 09:24, Dang, Christophe a écrit :
> Hello,
>
> I implemented the Fitzgibbon algorithm[1] for ellipse fitting. This algorithm extracts the eigenvalues and -vectors from a matrix, and therefore uses the spec function.
>
> When I test the eigenvalue, e.g.
> eigval > 0
> I get an error message 144 "Undefined operation for the given operands."
>
> I found a workaround
> sign(eigval) == 1
>
> But would like to understand why I get this message.
The result of the spec function is a vector of complex number unless the 
matrix is exactly symmetric, so the test

eigval > 0

has no meaning. May be you want to test :

real(eigval)>0

> Did I do something wrong?
> Is this a documented limitation?
> Is this a bug?
>
> Thanks in advance
>
> [1] Halir, R. and Flusser, J., Numerically Stable Direct Least Squaes Fitting of Ellipses, in Winter School of Computer Graphics, vol. 6, 1998
> http://wscg.zcu.cz/wscg1998/papers98/Halir_98.ps.gz
>