One texton vocabulary, learnt from the whole scene, and the map colours every pixel by its nearest texton — grass, bark, fur, foliage and sky fall into different textons. Drag the two boxes onto any pair of textures and read their histograms and χ² separation. Drop a whole bank — edges, bars, or blobs — or a single filter, and change k: how few filters and how small a vocabulary still keep χ² high?