image → image
Convolves the image with an isotropic 2-D Gaussian kernel — the canonical linear low-pass filter. It suppresses high-frequency noise while preserving overall structure, at the cost of blurring edges. Being separable and scale-space optimal, it is the usual first step before gradient, edge or threshold operators.
| a | kernel sigma (blur radius). σ ≈ 0.3 + 2.7·a — small a = gentle denoise, large a = strong blur. |
| b | reserved / unused for this operator. |
Denoise first, then differentiate/segment. Over-blurring (large a) erases fine detail, so pick the smallest sigma that removes the noise you care about.
gaussian 0.4 0.5 sobel_mag 0.5 0.5 otsu 0.5 0.5
▸ Load this pipeline · Load & run
median · bilateral · mean_box · sobel_mag
Provenance: separable Gaussian convolution (scipy.ndimage). See docs/PROVENANCE.md.