In 1966, Professor Seymour Papert at MIT conceptualized an ambitious summer project titled The Summer Vision Project. The task for the graduate student was to plug a camera into a computer and enable it to understand what it sees! I am sure it would have been super-difficult for the graduate student to have finished this project, as even today the task remains half complete. 

A human being, when they look outside, is able to recognize the objects that they see. Without thinking, they are able to classify a cat as a cat, a dog as a dog, a plant as a plant, an animal as an animal—this is happening because the human brain draws knowledge from its extensive prelearned database. After all, as human beings, we have millions of years' worth of evolutionary context that enables us draw inferences from the thing that we see. Computer vision deals with replicating the human vision processes so as to pass them on to machines and automate them.

This chapter is all about learning the theory and implementation of computer vision through machine learning (ML). We will build a feedforward deep learning network and LeNet to enable handwritten digit recognition. We will also build a project that uses a pretrained Inception-BatchNorm network to identify objects in an image. We will cover the following topics as we progress in this chapter:

• Understanding computer vision
• Achieving computer vision with deep learning
• Introduction to the MNIST dataset
• Implementing a deep learning network for handwritten digit recognition
• Implementing computer vision with pretrained models