Hands-On Image Processing with Python

HOG training

The SVM trainer selects the best hyperplane to separate positive and negative examples from the training set. These block descriptors are concatenated, converted into the input format for the SVM trainer, and labelled appropriately as positive or negative. The trainer typically outputs a set of support vectors—that is, examples from the training set that best describe the hyperplane. The hyperplane is the learned decision boundary separating positive examples from negative examples. These support vectors are used later by the SVM model to classify a HOG-descriptor block from a test image to detect the presence/absence of an object.