AI FOR VET • MODULE 7

Computer Vision Foundations: 2D Convolutional Kernels

How Convolutional Neural Networks (CNN) extract edges, textures and features from images.

🎯 Learning Outcomes / Ishodi učenja:
  • Understand a digital image as a 2D matrix of pixel brightness values.
  • Calculate 3×3 convolution kernel dot products step-by-step.
  • Apply smoothing/blur, sharpening, and edge detection (Sobel, Laplacian).
  • Inspect the mathematical calculation by tapping any pixel in the output.

A convolution kernel is a small 3×3 magnifying glass sliding over every image pixel. It multiplies the 9 neighboring pixels by its weights and sums them up. All modern vision AI (face recognition, defect detection, autonomous cars) begins with convolution.

S(i, j) = ∑ ∑ I(i − m, j − n) · K(m, n)

Tap any pixel on the result canvas to reveal its 9 inputs, weights, and divisor.

🔬 Interactive Laboratory / Interaktivna laboratorija:

🔍 Guided Investigation / Istraživački zadatak:
  1. Interactive inspector: Click any pixel in the output image to see the full formula and calculation.
  2. Edge detection: Select preset Sobel horizontal or Sobel vertical. How do edges highlight?
  3. Gaussian blur: Switch image to Noisy and select Gaussian blur. Watch noise disappear.