Pattern-based Simulation using Self Organized Maps

Abstract

An automatic cluster number selection algorithm for pattern-based simulation is presented in this paper. Patterns are generated from training images using a template. The generated patterns are classified into different classes using the self organised maps (SOM). The self organised maps are applied to classify the patterns by projecting the high-dimensional data to the two-dimensional lattice. The cluster numbers are automatically selected based on the U-matrix generated by the SOM. The classes are represented by a class prototype. The simulation was performed by measuring the similarity of the conditioning data event with the class prototypes using the L2-norm. The method is validated by a number of examples of conditional and unconditional simulations. The results show that the spatial continuity is well reproduced in all examples of conditional as well as unconditional simulations. The results were then compared with filtersim technique; and results revealed that the proposed algorithm performed better than the filtersim in all examples presented in this paper.