I've been reading on SVM and I learned that it doesn't support natively multiclass classification but it usually uses two different methods,
-one-vs-all, that generate a classifier for each class against the rest,
-one-vs-one, that considers classes two by two and makes a classifier for each pair.
What I'm confused about is the criteria that it's used to assign the class once the results of all the classifiers are out. I can understand taking a majority vote in the OvO case (although I have no idea what it's done in case of a draw) but I have no idea what's the solution for OvA.
Pic related is what causes me most issue since elements in the outside corners will get two matches (which is a draw again) but an element in the center triangle wouldn't be a match for any of the classifiers. In this case does SVM reject the element as not belonging to any of the classes or is there a catch?