2D Convolution

Hover over the images to see the kernel sliding over the 2D input. Same flip-and-slide, now in two dimensions.

b2slab.upc.edu ↗
Input image
Kernel
Size 3x3
Noise on input
Level 0
Kernel values (drag to tilt, dblclick to reset)
-1 +1
3x3 box
Equation
(f * g)(i,j) = Σm Σn f(i+m, j+n) · g(m, n)
Slide the kernel over every pixel,
multiply element-wise, sum.
Input f(i,j) hover to position kernel
Output (f * g)(i,j) convolution result
64 x 64 | hover to inspect
convolution with current kernel
Patch ⊙ Kernel = Product → Σ = output pixel position: (--,--)
f patch
⊙
g kernel
=
product
→
Σ all -- = output(--,--)