Edge Detection Visualizer | Canny Sobel Prewitt Image Processing

What is Convolution?

A kernel (filter) slides across an image. At each position, we multiply overlapping pixel values with kernel weights and sum the results — producing one output value per position.

What is a Kernel?

A small matrix (usually 3×3) of weights. Different kernels detect different features: edges, blurs, sharpness. The weights determine what the filter "looks for".

Padding

Valid: no padding — output shrinks. The kernel stays fully inside the image.
Same: zero-pad borders so output matches input size. Useful in deep networks.

Stride

How many pixels the kernel moves each step. Stride 1 = overlapping windows (dense output). Stride 2 = skip every other pixel (half-size output, faster).

Feature Map

The output grid after convolution. Each pixel represents how strongly the kernel's feature (e.g. a vertical edge) appeared at that location in the input.

Convolution Formula

(I ★ K)[i,j] = Σ_mΣ_n I[i+m, j+n] · K[m,n]

Where I is the input image and K is the kernel. Each output pixel is a dot product of the kernel with the local input patch.

Sobel Operators

G = √(Gx² + Gy²)

Sobel X detects horizontal intensity gradients (vertical edges). Sobel Y detects vertical gradients. Magnitude combines both for full edge strength.

Output Dimensions

out = ⌊(n + 2p − k) / s⌋ + 1

n = input size, p = padding, k = kernel size, s = stride. Try changing padding and stride to see the output grid resize.

In Deep Learning

CNNs (Convolutional Neural Networks) stack many such convolution layers. Each layer learns its own kernel weights automatically — discovering edges, textures, and objects.

Ready — hover over input image to preview

Input Image 20 × 20

Active Kernel Sobel X

Calculation

Hover or step to view…

Output size —

Progress 0 / 0

Feature Map —

Input pixel × kernel weight

Active kernel region

Active output pixel

Computed output