Multi-feature texture image segmentation based on tessellation technique

Due to the complexity and diversity of texture image, traditional single feature based texture image segmentation method can not meet the requirement of segmentation accuracy. In order to improve the segmentation accuracy, this paper proposes a multiple features based texture image segmentation method that combines tessellation. Firstly, multiple texture features are defined according to the spatial correlation of the pixel greyscale. Then, the image region is divided into sub-regions with tessellation, and the homogeneous regions to be segmented are obtained from fitting these sub-regions. Furthermore, the global potential energy function is defined through defining the heterogenous potential energy function among the homogeneous regions of multiple feature images and the potential energy function describing the neighborhood relationship of the sub-regions. The non-constrained Gibbs probability distribution is built to construct a posterior distribution from which a texture segmentation model is established. Finally, the M-H (metropolis-hastings) algorithm is used to sample the posterior probability distribution, and the optimal image segmentation result is achieved based on the maximum a posterior (MAP) estimation. The segmentation experiments on the simulated texture images, remote sensing images, natural texture images as well as SAR (synthetic aperture radar, SAR) sea ice images were conducted, and the segmentation results were analyzed and compared with those obtained using the single feature method; the qualitative and quantitative test results verify the effectiveness of the proposed algorithm. © 2015, Science Press. All right reserved.