Research on Classification of Wood Surface Texture based on Markov Random Field

To classify wood by surface texture, rank-2 Gibbs-MRF model was established at first. Rank-2 Gibbs-MRF parameters (beta2~beta9) of 300 wood texture samples were estimated by least square method. Trough data analysis, conclusions were drawn as follow: parameters of different sort of wood texture present a scattered distribution; every Gibbs-MRF parameter denotes the intensity of certain texture cluster type; the more delicate the texture is, the larger the corresponding parameter is, contrarily the smaller. Parameter beta2 of radial texture is the maximum one, and beta2~beta5>0, beta6~beta9<0; beta2 of tangential texture is the minimum one, and beta2<0, beta3~beta9>0; beta6~beta9 of tangential texture is greater than that of radial texture. According to the separable criterion value of Gibbs-MRF parameters, seven parameters (beta2~beta7, beta9) were selected by Simulated Annealing Algorithm as inputs to BP Neural Networks to classify six sorts of wood texture. As a whole, the correct ratio of classification reaches 84.5%. Conclusions: the seven Gibbs-MRF parameters are valid to describe wood texture feature; it is feasible to classify wood by surface texture according to the seven parameters.