Every day, millions of people use filters to stylize an Instagram post, or Gaussian blur to edit a photo. Have you ever wondered what's really going on when you do this? It all comes back to one fundamental process.
Computerphile is one of my favorite YouTube channels. Their videos give engaging, concise, and enlightening explanations of the technology, science, math, and techniques that make modern life possible. In this video, they examine kernel convolution, a cornerstone of modern computational imaging and the mathematical concept that enables a wide range of capabilities, including sharpening, blurring, noise removal, and edge detection, all through one simple idea: adjusting the relationship of a pixel to its neighboring pixels.
When you perform a technique such as sharpening, in essence, you are simply restructuring that fundamental relationship. The beauty of this concept is its simplicity and accessibility that nonetheless begets powerful and far-reaching capabilities. At its core, kernel convolution is nothing more than choosing to add and multiply numbers in carefully chosen ways. When you process an image via one of the aforementioned techniques, you are essentially telling Instagram, Photoshop, or whatever software you're using to apply a specific kernel to the image and output a new image. Check out the video for a great explanation and demonstration of the concept!