What means NaN when using minMaxLox

edit

unfortunately, nanand ìnfare the result of "undefined behaviour", and will just get propagated as is.

// let's make one :)
float inf = 1.0/0.0;
cerr << inf << endl;
float nan = 0.0*inf;
cerr << nan << endl;

from here:

General note about OpenCV implementations: Using of NaN values

will provide undefined behavior and should be avoided if there is no special NaN-related note in function/algorithm documentation.

to find out, if your Mat contains NaN's, compare it to itself:

 Mat m = ...
 Mat mask = (m != m);
 int numNaNs = countNonZero(mask);

if you find any, you should remove them, using patchNaNs() before the next step of your algorithm.