A deep dive into the Kuwahara filter, originally developed in the 1970s for denoising medical imagery, and its evolution into a stylized rendering technique for games. Covers the original filter's mechanics (dividing a kernel into sectors, picking the lowest-variance sector), its limitations (box artifacts, noise susceptibility), and two major extensions: the generalized Kuwahara filter (circular kernel, 8 sectors, Gaussian/polynomial weights, weighted color blending) and the anisotropic Kuwahara filter (structure tensor, eigenvectors for directional kernel adaptation). Performance optimizations using polynomial approximations replace expensive Gaussian functions. Practical demonstrations show the effect applied to Final Fantasy XIV and combined with dithering for painterly visuals.
Sort: