multiclass svm, one vs all

Dear all I am trying to train a multiclass svm using one vs all method. I need some hints doing this. How should I define the reject class for each binary classifier? for example, if I want my first binary classifier to label one group as '1' and the rest as 'not1', then what could be the feature vector for the class 'not1'? should it be the average of the other classes' feature vectors?

thank you all