This section shows how convolutional layers work in greater depth. At a basic level, convolutional layers are nothing more than a set of filters. When you look at images while wearing glasses with a red tint, everything appears to have a red hue. Now, imagine if these glasses consisted of different tints embedded within them, maybe a red tint with one or more horizontal green tints. If you had such a pair of glasses, the effect would be to highlight certain aspects of the scene in front of you. Any part of the scene that had a green horizontal line would become more focused.
Convolutional layers apply a selection of patches (or convolutions) over the previous layer’s output. For example, for a face recognition task, the first layer’s patches identify basic features in the image, for example, an edge or a diagonal line. The patches are moved across...
