Image Processing

Now that we know how a digital image is represented, let's discuss how computers can use this information to find patterns that will be used to classify an image or localize objects. So, in order to get any useful or actionable information from an image, a computer has to resolve an image into a recognizable or known pattern. As for any machine learning algorithm, computer vision needs some features in order to learn patterns.

Unlike structured data, where each feature is well defined in advance and stored in separate columns, images don't follow any specific pattern. It is impossible to say, for instance, that the third line will always contain the eye of an animal or that the bottom left corner will always represent a red, round-shaped object. Images can be of anything and don't follow any structure. This is why they are considered to be unstructured data.

However, images do contain features. They contain different shapes (lines, circles, rectangles...