Computer vision starts with one small grid of numbers.

A 3×3 kernel slides over every pixel. At each pixel it multiplies and adds. Blur, sharpening and edge detection all use this one step. The first layers of a CNN use it too. Pick a kernel and hover over the output to see the sum for that pixel.

Input $f$

∗
=

Output $g = f * h$

Hover or tap the output image.

How to use this site

  1. Learn in order. Each lecture uses the one before it. The order is: pixels, filters, features, grouping, geometry, recognition.
  2. Work every example by hand before you open the solution. The endsem numericals are of the same type (Otsu, morphology, Hough, hierarchical clustering, CNN output sizes, IoU).
  3. Do the self-test at the bottom of each page, then mark the lecture done. The browser saves your progress.
  4. The night before the exam, read only the Recap boxes and the formula cheatsheet.

The course map