Laplace¶
A second-order, direction-independent (isotropic) derivative filter: the sum of second partial derivatives in x and y. Where Sobel/Scharr respond to the slope of an intensity change, Laplace responds to how that slope itself is changing, so it peaks right at edges and zero-crosses through them — useful for precise edge localization — but it’s also considerably more sensitive to noise than a first-order filter, since differentiating twice amplifies high-frequency variation. In practice it’s usually computed on an already Gaussian-smoothed image (Laplacian of Gaussian).
See also
Gaussian — the smoothing step Laplace is normally combined with
Feature Extraction — every other feature