Machine Learning for OpenCV 4 - Second Edition

Multiclass classification using SVMs

SVMs are inherently two-class classifiers. In particular, the most prevalent method of multi-class classification in practice has been to create |C| one-versus-rest classifiers (commonly referred to as one-versus-all (OVA) classification) where |C| is the number of classes and to choose the class that classifies the test datum with the highest margin. Another approach is to develop a set of one-versus-one classifiers and to select the class that is chosen by the most classifiers. While this involves building |C|(|C| - 1)/2 classifiers, the time for training classifiers may decrease, since the training data set for each classifier is much smaller.

Now let's quickly jump onto how you can apply multi-class ...