In this chapter, the conventional model-based synthesis is reviewed, followed by the description of a new algorithm derived from the structural identification of a generic MGRF model for fast texture synthesis . A comparison with other texture synthesis methods are also presented.
Conventional identification of a generic MGRF model via stochastic approximation involves simulating the stochastic generative process of a texture by a Markov chain. Each state of the chain is considered a realisation or a sample randomly drawn from the model distribution. Such an identification also synthesises texture. The structural identification of a generic MGRF model yields a texture description in terms of characteristic texels and their placement rules. The texel-based texture description leads to a fast texture synthesis method, called bunch sampling in [40]. To synthesise a new texture, the bunch sampling algorithm places texels, randomly sampled from the training texture, into a new image in line with the estimated placement rules.