Joint quantisation and error diffusion of colour images using competitive learning

A competitive learning scheme for colour image quantisation is elaborated, in which the dithering process to eliminate contouring effects is embedded in the quantisation process instead of performed a posteriori. Quantisation is performed by clustering in colour space. The dithering process is a simple error diffusion, in which the quantisation error made by one pixel is diffused to its local neighbourhood. An objective function which takes the dithering process into account is optimised by use of a competitive learning approach. In this way, the colour quantisation process is optimally adapted to the dithered image, and the dithering process is optimally adapted to the colour palette. For small colour palettes, this is demonstrated to improve the visual quality of quantised images