Pyramid-based texture synthesis [Heeger and Bergen] [46] is a very first `fast' algorithm that generates a new texture by matching certain statistics with the training sample. In the method, a texture is decomposed into pyramid-based representations, i.e. an analysis and a synthesis pyramids are constructed from the training and the output textures respectively. The algorithm iteratively refines a synthesised texture by matching its histogram against that of the training texture, and by matching the histogram of each level of the synthesis pyramid with that of the corresponding level of the analysis pyramid. The synthesis process involves mainly image pyramid operations and histogram matching.