Probabilistic mixture-based image modelling

During the last decade we have introduced probabilistic mixture models into image modelling area, which present highly atypical and extremely demanding applications for these models. This difficulty arises from the necessity to model tens thousands correlated data simultaneously and to reliably learn such unusually complex mixture models. Presented paper surveys these novel generative colour image models based on multivariate discrete, Gaussian or Bernoulli mixtures, respectively and demonstrates their major advantages and drawbacks on texture modelling applications. Our mixture models are restricted to represent two-dimensional visual information. Thus a measured 3D multispectral texture is spectrally factorized and corresponding multivariate mixture models are further learned from single orthogonal mono-spectral components and used to synthesise and enlarge these mono-spectral factor components. Texture synthesis is based on easy computation of arbitrary conditional distributions from the model. Finally single synthesised mono-spectral texture planes are transformed into the required synthetic multispectral texture. Such models can easily serve not only for texture enlargement but also for segmentation, restoration, and retrieval or to model single factors in unusually complex seven dimensional Bidirectional Texture Function (BTF) space models. The strengths and weaknesses of the presented discrete, Gaussian or Bernoulli mixture based approaches are demonstrated on several colour texture examples.