Edge detection (Canny) in Computer Vision - Model Pipeline Trace

Model Pipeline - Edge detection (Canny)

The Canny edge detection pipeline finds edges in images by smoothing, finding gradients, and selecting strong edges. It helps computers see outlines like we do.

Data Flow - 6 Stages

1Input Image

1 image x 256 height x 256 width x 1 channel (grayscale)→Original grayscale image loaded→1 image x 256 x 256 x 1

A photo of a cat in grayscale

↓

2Gaussian Blur

1 image x 256 x 256 x 1→Smooth image to reduce noise using Gaussian filter→1 image x 256 x 256 x 1

Blurred cat image with less sharp noise

↓

3Gradient Calculation

1 image x 256 x 256 x 1→Calculate intensity change (gradient magnitude and direction) using Sobel filters→1 image x 256 x 256 x 2 (magnitude and angle)

Gradient map showing edges' strength and direction

↓

4Non-Maximum Suppression

1 image x 256 x 256 x 2→Thin edges by keeping only local maxima in gradient direction→1 image x 256 x 256 x 1 (thinned edges)

Thin lines highlighting strongest edges

↓

5Double Threshold

1 image x 256 x 256 x 1→Classify edges as strong, weak, or non-edges using two thresholds→1 image x 256 x 256 x 1 (edge classification)

Strong edges in white, weak edges in gray, others black

↓

6Edge Tracking by Hysteresis

1 image x 256 x 256 x 1→Keep weak edges connected to strong edges, discard others→1 image x 256 x 256 x 1 (final edge map)

Clear edge outlines of the cat

Training Trace - Epoch by Epoch

Loss
0.5 |****
0.4 |***
0.3 |**
0.2 |*
0.1 | 
    +---------
    Epochs 1-5

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.45	0.60	Initial edge detection with many false edges
2	0.30	0.75	Better edge clarity after tuning thresholds
3	0.20	0.85	Edges are sharper and noise reduced
4	0.15	0.90	Strong edges well detected, weak edges properly filtered
5	0.12	0.92	Converged to stable edge detection performance

Prediction Trace - 6 Layers

Layer 1: Input Image

Layer 2: Gaussian Blur

Layer 3: Gradient Calculation

Layer 4: Non-Maximum Suppression

Layer 5: Double Threshold

Layer 6: Edge Tracking by Hysteresis

Model Quiz - 3 Questions

What is the main purpose of the Gaussian Blur step in Canny edge detection?

A To classify edges as strong or weak

B To find the gradient direction

C To reduce noise and smooth the image

D To thin the edges to one pixel

Key Insight

Canny edge detection works by carefully smoothing, finding gradients, and selecting edges with thresholds. This step-by-step process helps computers find clear edges like humans do, improving with tuning and iteration.