09. Learning to Classify Images

Classification Techniques

You’ve just seen a complete classification pipeline. For an image classifier, starting with some input images, computer vision techniques are used to process those images, and extract features, like distinguishing colors or shapes in that image.

Then, a classifier looks at these features and outputs a class, which is a label that describes the image.

**A classifier should predict that images with similar shapes or colors have the same class. **

We usually tell a classification model what to look for. For example, say we are looking at a bunch of images and we want to classify them into two classes: car and not-car!
To classify a car, we might write a program that looks for the different parts of a car: wheels, lights, windows, and so on, and then if those things are found, we’ll classify an image as a car. We decide what traits are important to look for.

Machine Learning

However, there's another way to create a classifier and that’s with machine learning.

Machine learning allows a computer to figure out things on its own by giving it lots of examples. So, instead of telling a model what traits to look for, with machine learning, we’d just give it lots of images of cars and not-cars and let it learn to recognize traits that differentiate them! It can learn to recognize wheels and windows and which classification algorithm is best for accurately classifying any given image as car or not-car!

Now, you might be wondering: how exactly does a model like this learn to classify different images?

Next, we’ll look at how machine learning techniques can actually be trained to classify sets of images.