Artificial texture generation using force directed self-organizing maps

The generation of artificial textures which describes regions that consist of uniform properties is a useful function in image synthesis systems. In this paper, a neural network model called FDSOM (force directed self-organizing map) is proposed to generate artificial textures with highly parallelism. Based on the equilibrium of interaction forces between nodes, the proposed model can generate various texture images with different characteristics such as granularity, inclination, and randomness. In this paper, many generated textures are presented with various characteristics. Moreover, the self-organization property for the texture generation is also investigated. Experiments show that the proposed model provides good results at a computational cost less than that of other approaches. Our future work is to apply this model to the texture segmentation