image → image
Computes the gradient magnitude with the Sobel operator: it convolves the image with a horizontal and a vertical 3×3 derivative kernel and returns √(Gx² + Gy²). Bright pixels mark strong intensity transitions — i.e. edges. The Sobel kernel combines differentiation with a little smoothing, so it is more noise-tolerant than a bare finite difference.
| a | output gain / normalisation of the magnitude. |
| b | reserved. |
Pre-smooth with gaussian to avoid amplifying noise, then follow with a threshold to turn the edge map into a region.
gaussian 0.3 0.5 sobel_mag 0.6 0.5 otsu 0.5 0.5
▸ Load this pipeline · Load & run
laplace · prewitt_mag · dog · canny · otsu
Provenance: Sobel 3×3 derivative kernels (numpy/scipy). See docs/PROVENANCE.md.