Image Pre-processing for Model Training Exercise

In Exercise 1, we experimented with some smoothing and with Otsu's method for some basic image preprocessing. There are other types of image pre-processing that become important when we start using deep learning. While pre-processing is classically thought of as being useful for cleaning up images by removing artifacts and smoothing out noise, there's another purpose for pre-processing that's more specific to deep learning. That purpose is to expand our set of training data, and it is accomplished by a process called image augmentation. This process takes a set of real images that you have selected for training and creates augmented copies of them to make the set larger, but also to expand the range and variety of the training data that your model will see.

In this exercise, you'll use Keras' built-in ImageDataGenerator class to perform augmentation on a set of mammography images to transform those images into a wider variety of appearances for training a future model. Note that there is an infinite number of ways that you can augment an image, and the ways in which you do this should reflect what's possible to see in the real world.