38. Selecting Features
Types of Features
We've described features as measurable pieces of data in an image that help distinguish between different classes of images.
There are two main types of features:
- Color-based and
- Shape-based
Both of these are useful in different cases and they are often powerful together. For example, say I wanted to classify a stop sign. Stop signs are supposed to stand out in color and shape! A stop sign is an octagon (it has 8 flat sides) and it is very red. It's red color is often enough to distinguish it, but the sign can be obscured by trees or other artifacts and the shape ends up being important, too.
As a different example, say I want to avoid crashing into a car (a very important avoidance case!). I'll want to classify the object as a car, or at least recognize the car's boundaries, which are determined by shape. Specifically, I'll want to identify the edges of the vehicle, so that I can track the car and avoid it. Color is not very useful in this case, but shape is critical.
As you continue learning, keep in mind that selecting the right feature is an important computer vision task.