32. Features

Distinguishing and Measurable Traits

When you approach a classification challenge, you may ask yourself: how can I tell these images apart? What traits do these images have that differentiate them, and how can I write code to represent their differences? Adding on to that, how can I ignore irrelevant or overly similar parts of these images?

You may have thought about a number of distinguishing features: day images are much brighter, generally, than night images. Night images also have these really bright small spots, so the brightness over the whole image varies a lot more than the day images. There is a lot more of a gray/blue color palette in the day images.

There are lots of measurable traits that distinguish these images, and these measurable traits are referred to as features.

A feature a measurable component of an image or object that is, ideally, unique and recognizable under varying conditions - like under varying light or camera angle. And we’ll learn more about features soon.

Standardizing and Pre-processing

But we’re getting ahead of ourselves! To extract features from any image, we have to pre-process and standardize them!

Next we’ll take a look at the standardization steps we should take before we can consistently extract features.