The past – classical computer vision

In the past, programmers would go through a painstaking and long process to create computer vision algorithms that were, well, less than satisfactory. This process was called feature engineering.

Essentially, you'd manually design features or filters that would act as filters on an image, and if an image is activated above a certain threshold by certain, handcrafted filters, it would be given a certain class.

This technique is not very scalable, takes a very long time, takes smart people, and doesn't provide very good accuracy, at least by today's standards.