Nonparametric Markov random field order estimation and its application in texture synthesis

In this paper we propose a new theory for the order estimation of nonparametric Markov random field (N-MRF) model. Texture synthesis based on N-MRF model performs well visually for a wide range of natural textures, [9]. The result of texture synthesis is dependent upon the model order, and the computational complexity increases parabolicaly with the model order. Therefore, it is required to estimate the minimum model order for computationally efficient texture synthesis. In the proposed methodology, the basic definition of local conditional density is redefined. The proposed model order estimation (MOE) approach for N-MRF model has been tested with a number of stochastic and near-regular textures, collected from the Brodatz's standard database [3]. Results show the efficacy of the proposed approach in solving the MOE problem efficiently.