Image Processing: Why You Should Normalize Your Images

1. Normalizing Homographies
A homography represents a transformation between coordinates. But the actual values that go into this homography can be arbitrarily large and small. The reason to normalize here is to more accurately represent large numbers or numbers close to 0, due to computer system limitations on floats and propagation of precision errors.


2. Normalize Images
Normalizing images from their byte representations to a value between 0 and 1 is for convenience, in order to get a better estimate of the pixel intensity. We have better intuition towards a value that can be represented as a percentage (0 to 1) vs. a pixel value (0 to 255).


Source: Karanhaar