Edges and Segments

Build filters, find an automatic threshold and analyse connected regions.

Build the edge filter

Select a weight and use the stepper. True convolution flips the kernel horizontally and vertically before multiplying.

Threshold marker80

Selected kernel weight

ConnectivityConvolution mode

First try: 100 points

Concept: each filter output combines a local neighbourhood.

The maths

True convolution: g(y,x) = ΣΣ f(y+i,x+j)k(−i,−j).

Otsu: maximise w₀w₁(μ₀−μ₁)²; class 0 contains levels ≤ t. Choose the smallest threshold on ties.

Connected components: foreground pixels are non-zero. Diagonal neighbours connect only under 8-connectivity.

Assumption: grayscale pixels are integer levels from 0 to 255 for Otsu.