Dropout in Neural Network

We discussed the overfitting problem last time. Now let's understand the dropout from above plots. The plots above are showing models for classifying dogs and cats. As a human being, I can easily distinguish a cat and a dog, no matter it is wearing a mask or in red or not. So I will assume that in the above plot, the mask of the cat or the red cloth of the dog is a noise since those are not specific features for the animal species.