Image approximation via evolutionary search with geometric and colour constraints.
Afghan Girl (McCurry) 64 ellipses, 32-colour paletteNapoleon Crossing the Alps (David) 95 equilateral triangles, 6-colour paletteGreat Wave off Kanagawa (Hokusai) 120 rotated rectangles, 16-colour paletteStarry Night (van Gogh) 689 lines, 32-colour paletteComposition with Large Red Plane, Yellow, Black, Grey and Blue (Mondrian) 23 rectangles, 6-colour paletteWanderer above the Sea of Fog (Friedrich) 49 quadrilaterals, 1-colour tintCat with a Pearl Earring (Herbert) 65 triangles, 32-colour paletteWater Lily Pond (Monet) 448 lines, 16-colour paletteTower of Babel (Bruegel) 127 pentagons, 4-colour tintMona Lisa (da Vinci) 48 ellipses, 1-colour tintMona Lisa (da Vinci) 60 equilateral triangles, 2-colour tintMona Lisa (da Vinci) 49 equilateral triangles, 3-colour tintMona Lisa (da Vinci) 54 equilateral triangles, 4-colour tint
This page presents research outputs from evolutionary image approximation experiments conducted for academic and pedagogical purposes.
The displayed SVGs are lossy vector reconstructions generated through optimization against reference images and are used to study approximation efficiency, representation sparsity, and search dynamics.
No machine learning model is trained, and no generative AI or style synthesis is involved.