Edge-preserving smoothing via nonlinear diffusion
Perona and Malik (1990) introduced anisotropic diffusion as a technique for edge-preserving smoothing. Unlike isotropic diffusion (Gaussian blur), this method adapts the diffusion rate based on local image gradients.
Kappa (κ): Controls edge sensitivity. Higher values allow more
diffusion across edges. Lower values preserve more edges but may leave noise.
Time Step (dt): Must be small (≤0.25) for stability.
Iterations: More iterations = more smoothing.
The conductance function c(|∇I|) approaches 1 in flat regions (allowing diffusion) and approaches 0 at edges (stopping diffusion), creating edge-preserving behavior.