Recall that in previous chapters, we made the distinction that neural networks require numerical inputs. We saw how we can encode categorical features, such as day of week, into numerical features using one-hot encoding. How then do we use an image as input for our neural network? Well, the short answer is that all digital images are numerical in nature!
To see why this is so, consider a 28 x 28 image of a handwritten digit 3, as shown in the following screenshot. Let's assume for now that the image is in grayscale (black and white). If we look at the intensity of each pixel that makes up the image, we can see that certain pixels are totally white, while some pixels are gray and black. In a computer, white pixels are represented with the value 0 and black pixels are represented with a value of 255. Everything else in between white and...
