Perlin Noise is a set of data, which is procedural, deterministic, seeded, and can be expanded to an arbitrary number of dimensions.
Procedural - The data is generated, meaning it was not created manually by a person. It is generated using code.
Deterministic - Although generated through code and is seemingly random, the data can be repeatable if the starting conditions are the same. This means you can generate the same set of values again if you want to - using a 'seed'.
Seeded - This is what actually makes the randomness deterministic. You can provide a ‘seed’ - usually a number of some kind, which will then be used to generate the noise. If the same number is used again in the future, the exact same set of numbers will be generated!
Dimensions - The dimensions basically mean how many directions are the numbers generated in. In 1-D the numbers can go left and right (like a number line). In 2-D the numbers can go left, right, up and down (like a number for every point on a flat surface). Perlin noise however can be applied to any number of dimensions; 3-D, 4-D, even 100-D if you wanted to!
Have a look at what raw Perlin Noise looks like below! This is noise in 2 dimensions (x and y dimension), where each value is mapped to an intensity for the pixel (where large numbers are close to white, and small numbers are close to black).
There are four main steps to generating Perlin Noise, which must be applied one after the other. Some different implementations might do some parts slightly differently, depending on the use case of the noise, but we will talk about these later.
The main parts are:
1: Generating Grid Vectors
2: Creating Candidate Point Vectors
3: Calculating Dot Products
4: Interpolation