Representing images

One of the most common and important data types for computer vision is, of course, images. The most straightforward way to represent images is probably by using the grayscale value of each pixel in the image. Usually, grayscale values are not very indicative of the data they describe. For example, if we saw a single pixel with a grayscale value of 128, could we tell what object this pixel belonged to? Probably not. Therefore, grayscale values are not very effective image features.

Using color spaces

Alternatively, we might find that colors contain some information that raw grayscale values cannot capture. Most often, images come in the conventional RGB color space, where every pixel in the image gets an...