Mastering Computer Vision with TensorFlow 2.x

Running AdaBoost training

The image is divided into T windows where Haar-like features are applied and their value is calculated as described previously. AdaBoost builds a strong classifier from a large number of weak classifiers by iterating over a training set of T windows. At each iteration, the weights of the weak classifier are adjusted based on a number of positive samples (faces) and a number of negative samples (non-faces) to evaluate the number of misclassified items. Then, for the next iteration, the weights of the misclassified item are assigned a higher weight to increase the likelihood of these being detected. The final strong classifier h(x) is a combination of weak classifiers weighted according to their error.

• Weak classifier ...