Voronoi tessellation‐based regionalised segmentation for colour texture image

This study presents a region‐based algorithm for segmenting colour texture image, which uses Voronoi tessellation for partitioning the domain of the image and Markov random field (MRF) for modelling colour texture. In detail, (i) an image domain is divided into polygons (or sub‐regions) by Voronoi tessellation; (ii) two MRF models, improved Potts model and multivariate Gaussian MRF model, are used to characterise colour texture structures inter‐ and intra‐polygons, respectively; (iii) by Bayesian paradigm, a posterior distribution which characterises the segmentation and model parameters conditional on a given colour image can be obtained up to a normalising constant; (iv) a Markov chain Monte Carlo algorithm is developed to simulate from the posterior distribution; finally, (v) a maximum a posteriori scheme is employed to find an optimal segmentation and model parameters. In order to evaluate the proposed colour texture segmentation algorithm, two kinds of colour texture images are tested, including synthetic and real colour texture images. The accuracy assessments are performed qualitatively on all kinds of images and quantitatively on synthetic images. All results demonstrate that the proposed algorithm is efficiently.