Feature extraction is about finding a dense representation of the input.

Given a set of features, feature selection picks a subset of the features relevant in making a prediction.