CS5720 - Week 4
Slide 64 of 80

The Convolution Operation: Intuitive Explanation

What is Convolution?

Convolution is a mathematical operation that slides a small matrix (filter/kernel) across a larger matrix (image), computing element-wise multiplications and sums at each position.
Key Properties:

Local Connectivity: Each output depends only on a small region
Weight Sharing: Same filter applied everywhere
Translation Equivariance: Shifted input → shifted output
Feature Detection: Filters learn to detect patterns
The Formula:
Output(i,j) = Σ Σ Input(i+m, j+n) × Kernel(m,n)

Real-World Analogies

🔍 The Magnifying Glass
Sliding a magnifying glass over a document to examine each part - the lens is your kernel!
📌 The Pattern Stamp
Like using a stamp to check if a pattern matches at each position on a page.
🪟 The Sliding Window
Looking through a small window that slides across a large wall, seeing only what's visible through the window.
🔦 The Flashlight Beam
Scanning a dark room with a flashlight - you process the room one illuminated patch at a time.

Interactive Convolution Demonstration

Input (5×5)
×
Kernel (3×3)
=
Output (3×3)
Current Calculation: Ready to start
Prepared by Dr. Gorkem Kar