OpenCV 4 with Python Blueprints - Second Edition

Using SVMs for multiclass classification

Whereas some classification algorithms, such as neural networks, naturally lend themselves to using more than two classes, SVMs are binary classifiers by nature. They can, however, be turned into multiclass classifiers.

Here, we will consider two different strategies:

• One-versus-all: The one-versus-all strategy involves training a single classifier per class, with the samples of that class as positive samples and all other samples as negatives.

For the k classes, this strategy thus requires the training of k number of different SVMs. During testing, all classifiers can express a +1 vote by predicting that an unseen sample belongs to their class.

In the end, an unseen sample is classified by the ensemble ...